Intelligent Weed Detection using Machine Learning

Intelligent Weed Detection using Machine Learning

Walid Lekheouayene, ArunaDevi Karuppasamy*

Department of Computing Sciences, Gulf College, Sultanate of Oman, Muscat

*Corresponding Author

DOI : https://doi.org/10.51244/IJRSI.2024.1114SDAICO7

Received: 04 July 2024; Accepted: 16 July 2024; Published: 02 November 2024

ABSTRACT

The study aims to create a sophisticated weed identification system that utilizes machine learning algorithms. Using image analysis techniques, the main goal is to develop and execute a model that can effectively recognize and differentiate crops and weeds. These models will form the basis for an automated weeding system, allowing for accurate and selective weed removal while minimizing the need for physical labour and harmful chemical herbicides. This study will encompass crucial stages, such as data collecting, preprocessing, model training, and deployment. Initially, an extensive dataset of images illustrating crops and diverse weed species will be gathered and organized. The dataset will be subjected to preparation procedures, including image resizing, augmentation, and normalization, to ensure consistency and improve the training process. Deep learning approaches such as convolutional neural networks (CNNs), will be utilized for model training and optimization. These algorithms will acquire the ability to identify and isolate essential characteristics from the image data, facilitating precise classification of crops and weeds. After completing the training and validation process, the created models will be included in an automated weeding system. This System will utilize computer vision techniques to analyses field images in real-time and detect instances of weed infestations. According to the model’s predictions, specific and focused methods of removing weeds can be used, eliminating the need for using herbicides without discrimination and reducing the unintended harm to crops.

Keywords: Weed Detection, Machine Learning, Convolutional Neural Network.

INTRODUCTION

Weeds pose a severe threat to agricultural productivity, resulting in substantial reductions in crop yield. Traditional methods of controlling weeds, such as manual removal and haphazard use of chemical herbicides, require much labour, take up a significant amount of time, and are susceptible to human mistakes. In addition, the overuse of herbicides can cause damage to plants that were not intended to be targeted and can also contribute to the deterioration of the (Bah, Hafiane and Canals, 2018). Precision agriculture, a method that entails the precise application of herbicides in the appropriate location and timing, has been suggested as a more sustainable and efficient strategy for weed control (Bah, Hafiane and Canals, 2018). There is an urgent need to develop automated weed management systems that utilize artificial intelligence (AI) and machine learning (ML) to overcome the limits of conventional weed control approaches. By harnessing this cutting-edge technology, it is feasible to offer a more accurate, effective, and eco-friendly resolution to the weed control obstacles farmers encounter. Researching the application of machine learning to identify and treat weeds is of great importance. This approach allows for precise weed detection and tailored treatment, reducing herbicide need and minimizing the adverse environmental effects.

Conventional methods of controlling weeds, including physical removal and the use of chemical herbicides, are now widely acknowledged as unsustainable and harmful to the environment. Manual weeding requires a significant amount of labour, takes up much time, and is susceptible to human mistakes. On the other hand, the uncontrolled application of chemical herbicides can damage plants that are not the intended target, contaminate soil and water sources, and contribute to the deterioration of the environment. Incorporating artificial intelligence (AI) and machine learning (ML) technology offers the potential to transform weed control practices by implementing automated and highly efficient solutions.

This project intends to create an accurate and targeted weed detection and elimination system using sophisticated image analysis and machine learning methods. This strategy can potentially decrease the dependence on manual labour, reduce the utilization of toxic chemical herbicides, and ultimately encourage sustainable agricultural practices. Moreover, the effective execution of this project can have extensive ramifications for various stakeholders within the agriculture sector. Agricultural practitioners can benefit from enhanced agricultural productivity, fewer workforce expenses, and mitigating environmental repercussions. Consumers can experience superior-quality produce grown using sustainable methods, while the environment can gain from decreased chemical runoff and enhanced resource utilization. This study aims to tackle weed control difficulties by combining artificial intelligence (AI) and machine learning (ML) technology. It has the potential to enhance the agricultural business, promote food security, preserve the environment, and drive economic growth in the sector.

Utilizing machine learning for intelligent weed detection has great potential for tackling the difficulties linked to traditional weed management methods. Nevertheless, other unresolved concerns and gaps still need to be dealt with. A major obstacle is in the proper differentiation of crops and weeds since they have striking visual resemblances, particularly during their first phases of growth (Bah, Hafiane and Canals, 2018; Wang, Zhang and Wei, 2019). The identification procedure is further complicated by varied lighting conditions, overlapping plants, and different growth phases, exacerbating the difficulty. Another notable deficiency is the requirement for extensive, annotated datasets to train machine learning models with optimal effectiveness. Generating such datasets is demanding and time-consuming, typically involving the human annotation of many photos (Bah, Hafiane and Canals, 2018). Moreover, the existing method of uniformly applying herbicides leads to environmental degradation and diminishes field productivity (Pantazi, Moshou and Bravo, 2016; Liu, 2022) . To successfully use Intelligent Weed Detection systems in agricultural contexts, it is necessary to address these shortcomings by developing robust, flexible, and user-friendly AI/ML-based solutions.

These solutions must accurately distinguish between crops and weeds, considering various field conditions, and smoothly integrate with current farming procedures. Addressing these obstacles will not only improve the effectiveness of weed control but also encourage the adoption of sustainable farming methods and mitigate the ecological consequences of herbicide usage. The development of AI and ML-powered automated weed detection systems has significant consequences for stakeholders in the agricultural sector. The groups impacted include farmers, consumers, and the environment.

The main objective of this study is to create machine learning models that can effectively differentiate between crops and weeds using image analysis techniques. Integrating these models into automated weeding systems will allow for accurate and targeted weed removal, hence lowering the need for physical labour and minimizing the usage of chemical herbicides. This study seeks to utilize artificial intelligence (AI) and machine learning (ML) to tackle the obstacles associated with weed management. Its objectives include promoting sustainable agricultural practices, improving crop yields and quality, and contributing to the general progress of the agricultural business. This study aims to utilize the capabilities of artificial intelligence (AI) and machine learning (ML) to tackle a significant problem faced by farmers globally. The ultimate goal is to promote food security, preserve the environment, and foster economic growth in the agricultural industry. Machine learning algorithms for intelligent weed detection are essential for sustainable agriculture. They have the potential to help farmers, the environment, and consumers by enabling more precise and environmentally friendly weed management procedures.

LITERATURE REVIEW

This study aims to offer a thorough understanding of the theoretical foundations and related work in the field of machine learning-based weed detection. The paper will begin by exploring the theoretical foundations, including different machine learning algorithms and deep learning architectures used for weed identification. It will then examine the practical applications and implementations of these techniques, discussing the challenges, innovative approaches, and potential solutions put forth by researchers. This literature review aims to analyses the existing knowledge, identify research gaps, and explore future directions for the development of intelligent weed detection systems. The findings will help advance sustainable agriculture by facilitating efficient and eco-friendly weed management strategies.

Theoretical Review

Machine learning in agriculture is becoming more popular for finding and identifying weeds. Algorithms can use pictures or sensor information to tell the difference between crops and unwanted types of weeds. Now, will examine the fundamental ideas that make this field interesting. The Random Forest algorithm is highly popular and widely utilized. According to Gašparović et al. (2020), this ensemble learning method effectively differentiates weeds and bare soil, even when the training dataset sizes vary. Many researchers favour this option due to its strong and reliable performance. The Support Vector Machine (SVM) is a highly effective supervised learning model that has demonstrated its capabilities in various weed detection studies (Sarvini et al., 2019; Islam et al., 2021). SVMs can efficiently classify crops and weeds by identifying the optimal hyper-plane that separates the classes. The rise of deep learning, specifically Convolutional Neural Networks (CNNs), has been a significant game-changer. Artificial neural networks, which draw inspiration from the human visual cortex, have shown impressive accuracy in analyzing low-resolution weed images (Urmashev et al., 2021). YOLOv5 and other advanced CNN architectures are widely used for accurate weed categorisation. Transfer learning, a process known as transferring knowledge from pre-trained models like ResNet50 to weed detection tasks, has also been investigated by researchers (Ukaegbu et al., 2021). This method utilizes the feature extraction capabilities of these models, refining them for the problem at hand. Exploring new frontiers, certain studies have delved into unsupervised and one-class classification methods. These techniques have achieved impressive accuracy comparable to supervised models by utilizing deep features and unsupervised data (Bah, Hafiane and Canals, 2018). In related tasks such as defect detection, hybrid models that combine machine learning with optimization algorithms have shown promise. This suggests potential applications in weed identification (Ibrahim et al., 2022).

Related Work

Studies have examined the complex difficulties of distinguishing crops from weeds through image analysis. According to (Wang, Zhang and Wei, 2019), a major challenge is the striking resemblance between these plants in terms of their biological morphology and spectral features. Factors such as lighting conditions, growth stages, and visual appearance variations add to the complexity (Lottes et al., 2020). To address these challenges, researchers have utilized a wide range of techniques. Machine vision and image processing techniques on the ground have been extensively used for weed detection. These methods include pre-processing, segmentation, feature extraction, and classification (Wang, Zhang and Wei, 2019). Advanced machine learning algorithms like Convolutional Neural Networks (CNNs), Artificial Neural Networks (ANNs), and Support Vector Machines (SVMs) have been combined with data from spectroscopy, colour imaging, and hyper- spectral imaging to improve recognition capabilities (Su, 2020). A potential method involves using fully convolutional networks to estimate plant stem positions and conduct pixel-wise semantic segmentation (Lottes et al., 2020). These models can enhance their robustness and adapt to environmental variations using spatial information and crop arrangement from image sequences rather than single frames. Researchers have explored different avenues to address the challenge of acquiring high-quality datasets. Unmanned Aerial Vehicles (UAVs) have become a valuable tool for collecting high-resolution and adaptable data streams (Gallo et al., 2023). In addition, researchers have used procedural generation techniques to create synthetic training datasets. These techniques introduce random variations in elements such as crop species, soil types, and lighting conditions (Di Cicco et al., 2017). Specialized datasets, such as the CropAndWeed dataset (Steininger et al., 2023), have also been curated. This dataset offers detailed annotations for specific crop and weed species, including bounding boxes, semantic masks, and stem positions. In addition, researchers have explored active learning strategies as practical approaches for developing precise crop-weed classification systems. These strategies involve utilizing a small yet diverse subset of the available data (Yang et al., 2022). In recent studies, researchers have delved into unsupervised data labelling techniques. These techniques utilize automated crop row detection to identify inter-row weeds effectively. This identification aims to train convolutional neural networks, as discussed by Bah, Hafiane, and Canals in 2018. This method decreases the time-consuming task of manually annotating by experts. Many studies have focused on the development of automated weed management systems. Support Vector Machines (SVMs) have shown remarkable accuracy in classifying crops and weeds from digital images, reducing misclassification (Ahmed et al., 2012). Convolutional Neural Networks (CNNs) have been effectively incorporated into precision sprayers for site-specific weed management (SSWM), utilizing high-resolution UAV imagery (Khan et al., 2021).

Drones and advanced sensor technologies are now essential for effective weed management. When integrated with machine learning algorithms, these systems can identify weed patches and optimize their elimination through Autonomous Weeding Robot systems (Esposito et al., 2021). Active learning systems have shown high accuracy in distinguishing crops from weeds when using hyper- spectral sensing to identify different weed types (Pantazi, Moshou and Bravo, 2016). Agricultural researchers have made significant progress in creating robotic systems that can effectively remove weeds using mechanical and chemical techniques. These systems employ sophisticated algorithms to identify crops and weeds accurately, ensuring targeted and precise weed elimination (Wu et al., 2020). Researchers in large-scale production systems have also investigated nonselective weed management methods such as lasers and electrical weeding, specifically targeting individual weeds (Coleman et al., 2022). Machine learning algorithms integrated with UAV-based sensors have proven cost-effective in accurately identifying various weed species. This advancement paves the way for broader implementation of Sustainable Soil and Water Management (SSWM) processes (Etienne and Saraswat, 2019). Proposed for real-time control in specific crops like onions, computer vision-based robotic weed management systems combine image processing, machine learning, and IoT technologies (Arakeri et al., 2017). Researchers have taken diverse approaches in this field to address the challenges of accurate weed detection and management. The related work emphasizes the different strategies employed. These studies pave the way for developing intelligent and sustainable weed control systems, utilizing cutting-edge deep learning techniques and innovative data acquisition and labelling strategies.

Synthesis

There are many significant advantages to be gained from implementing intelligent weed detection systems. Farmers can benefit from these technologies as they help increase crop yields by precisely identifying and managing weeds. This prevents weeds from competing with crops for essential resources such as nutrients, water, and sunlight (Wu et al., 2020). Autonomous robotic weeding systems can potentially decrease the need for agrochemicals, which can help reduce pollution and promote sustainability. Reducing pesticide usage has clear benefits for consumers. It helps preserve agricultural ecosystems and promotes the production of healthier food products (Lin et al., 2017). Intelligent systems can enable site-specific weed management (SSWM) strategies, which offer economic and environmental benefits. These strategies focus control measures on areas where weed infestations surpass economic loss thresholds (Shaw, 2005). Intelligent weed detection systems align with the growing emphasis on sustainable agriculture. Robotic systems with cameras are increasingly used in the Indian agricultural sector to minimize the need for manual intervention with harmful chemicals and reduce the introduction of toxic substances into the food chain (Rani et al., 2022).

Accurate weed detection is crucial for precise spraying. It relies on precisely identifying and localizing weeds and crops (Wu et al., 2021). Utilizing unmanned aerial vehicles (UAVs) for weed identification in agricultural areas presents a novel and environmentally conscious method to improve weed surveillance (Mohidem et al., 2021). The progress made in mechanical weed control, particularly in targeting weeds within crop rows, has great potential. Weeders equipped with advanced sensors and robotics have the potential to reduce or eliminate the need for manual weeding significantly. This technology addresses environmental and food security issues while improving productivity and quality (Van Der Weide et al., 2008; Umamaheswari, Arjun and Meganathan, 2018). Many studies have focused on evaluating automated weed detection and management systems’ performance, efficiency, and environmental impact. One standard method is to use unmanned aerial vehicles (UAVs) to capture high-resolution images for analysis. In their study, Bah, Hafiane and Canals (2018) utilized convolutional neural networks (CNNs) and unsupervised data labelling to create a deep-learning approach for weed identification in line crops using UAV images. The method employed in this study successfully detected crop rows and utilized inter-row weeds as a training dataset for the CNNs, accurately identifying crops and weeds. The identification system made by Khan et al. (2021) uses deep learning to tell the difference between weeds and crops in strawberry and pea fields, aiming for its use in a sprayer for precision agriculture. Their system shows good accuracy, with an effective kappa coefficient, which agrees well with real data.

A strong or high kappa value (closer to 1) points towards good alignment between model forecasts and factual data – an understanding not simply due to luck. This becomes especially important in applications such as weed identification, where correct sorting is necessary for successful precision agriculture methods. In a comprehensive review by Su (2020), different sensing techniques were assessed to determine their effectiveness in distinguishing crops from weeds. The techniques examined included point spectroscopy, RGB, and hyper-spectral imaging. These methods were combined with machine learning algorithms such as convolutional neural networks (CNNs), artificial neural networks (ANNs), and support vector machines (SVMs), emphasising the significance of accurate plant identification for effective weed management. Parallel image processing has been suggested as a means to enhance the efficiency of weed detection systems. Umamaheswari, Arjun and Meganathan (2018) proposed using parallel processing on GPUs to achieve real-time image classification and weed detection in farm crops. This is an essential requirement for the effective implementation of IoT-based precision agriculture. Li et al. (2019) investigated machine vision-based plant detection technologies. They discussed different strategies to address the difficulties caused by changing light conditions and the requirement to differentiate between crops and weeds. They explored various approaches to address these challenges and emphasized the value of integrating conventional techniques to improve deep learning methods.

Researchers have also explored using computer vision and deep learning to detect weeds in real time. In their study, Junior and Ulson (2021) evaluated the effectiveness of systems utilising architectures such as YoloV5 in rapidly detecting resilient weed species. They concluded that these systems are well-suited for real-time applications. Olsen (2020) investigated the accuracy of identifying weed species in real-world settings The study utilized advanced technologies such as AutoWeed, which uses deep learning and machine vision to control weeds in agricultural fields and pastures efficiently. Liakos et al. (2018) conducted a comprehensive review of the application of machine learning in agriculture. They discussed various uses, such as yield prediction, disease detection, and weed detection. They also highlighted the transformation of farm management systems into real-time decision-support tools powered by AI. Firmansyah et al. (2022) proposed a technique to identify weeds in oil palm plantations in real time. By utilizing machine learning and image processing methods, they aimed to minimize environmental damage and boost productivity, thereby promoting environmental sustainability. In a recent study, Adeniji et al. (2023) presented an AI-driven robot designed for weed control in legume farms. The robot used advanced image recognition algorithms to identify and remove weeds, surpassing manual labour regarding speed and accuracy in field trials. Although there has been notable progress, the current literature also highlights gaps and areas that require further research. A common problem is the requirement for broader and more varied datasets to improve the accuracy of weed detection models in various situations (Murad et al., 2023).

Moreover, the fluctuating performance of machine learning models in different situations, like changes in image quality and lighting conditions, emphasizes the need for models that can withstand these variations (Urmashev et al., 2021). In addition, remote sensing technologies and deep learning algorithms for identifying weed growth have shown promise. However, more research and refinement are needed to effectively implement these methods in smart agriculture (Wang, 2023). Attention should be given to developing precise sensors and analysis methods to rapidly and non-destructively detect weeds within crop rows (Su, 2020). Further research should focus on optimizing deep learning models for real-time applications and integrating them with hardware accelerators and optimization approaches (Ukaegbu et al., 2021). Furthermore, the rise of hybrid machine-learning models and the requirement for comprehensive crop and weed imaging databases calls for enhanced collaboration among researchers and engineers (Liu and Bruch, 2020). The literature review provides an overview of the advancements in intelligent weed detection using machine learning. It also discusses the challenges and opportunities for further progress in this field. Researchers can pave the way for more efficient, sustainable, and intelligent weed management systems in agriculture by addressing current gaps and leveraging emerging technologies.

METHODOLOGY

The Intelligent Weed Detection Using Machine Learning project will use a hybrid research methodology that combines experimental and quantitative approaches. The researcher aims to integrate different methodologies to thoroughly investigate and develop an effective and comprehensive weed detection system. The experimental research will focus on designing, implementing, and evaluating machine learning models and algorithms to detect weeds. The researcher can use a practical approach to experiment with and improve various techniques within a controlled setting, including convolutional neural networks (CNNs), random forests, and support vector machines (SVMs). The researcher will train and validate these models during the experimental phase using meticulously selected datasets, including annotated images of crops and weeds in various field conditions. The quantitative research methodology will enhance the experimental component by offering a structured and data-driven approach to the project. Assessing the performance and accuracy of the developed models will heavily rely on statistical analysis. The researcher will use metrics like precision, recall, F1-score, and confusion matrices to measure how well the models can accurately identify and classify crops and weeds. In addition, quantitative methods are crucial for optimising model parameters, selecting features, and tuning hyperparameters to achieve the best possible performance. The hybrid research approach enables the exploration and evaluation of various machine learning algorithms, architectures, and techniques for weed detection.

The experimental component allows for exploring different deep learning models, including convolutional neural networks (CNNs) with varying architectures. In addition, we will also examine conventional algorithms such as random forests and support vector machines (SVMs) to assess how they perform compared to deep learning methods. An essential analysis aspect involves the quantitative component, which examines the models’ outputs, pixel values, and extracted image features. Analysing these factors statistically will help optimise model parameters, select features, and tune hyperparameters. This analysis will ensure that the models can accurately capture the visual patterns and characteristics needed to distinguish crops from weeds in different agricultural environments. The Intelligent Weed Detection Using Machine Learning project will significantly benefit from a comprehensive and well-rounded approach, combining experimental and quantitative research methodologies. The experimental part will drive the development and testing of innovative machine-learning models. In contrast, the quantitative part will provide a rigorous framework for evaluation, optimisation, and data-driven decision-making. This hybrid approach aims to enhance comprehension of the problem domain, foster the creation of reliable and precise models, and ultimately aid in developing an intelligent and efficient weed detection system for sustainable agriculture.

System Development Methodology (Agile)

The researcher plans to use the Agile methodology in developing the Intelligent Weed Detection Using Machine Learning system. The Agile approach is highly compatible with projects that require iterative development and continuous adaptation to changing requirements. This approach prioritises adaptability and the capacity to quickly incorporate feedback and evolve domain knowledge.

The Agile methodology consists of the following fundamental principles and practices:

1. Iterative and Incremental Development: The development process is broken down into brief iterations or sprints, usually lasting 2-4 Every iteration includes delivering a product increment that can be shipped, enabling ongoing feedback and adaptation.
2. Acquiring Cross-functional Expertise: The researcher will apply various skills and knowledge, including machine learning, data science, software development, and educational resources created by agricultural domain experts, such as videos and articles they authored.
3. Responding to Change: The Agile methodology promotes a flexible approach that welcomes change and encourages adaptation to new requirements or changes in domain knowledge. The researcher will consistently improve and optimise the system using feedback and insights from educational materials written by agricultural

The development process will follow an iterative cycle consisting of the following phases:

1. Planning: At the start of every iteration, the researcher will create a list of tasks and features to address. The planning process will be informed by the expertise in the field, which has been obtained from scholarly sources authored by agricultural
2. Analysis and Design: The researcher will carefully analyse the tasks and features in the backlog, incorporating the appropriate domain knowledge and adhering to the best practices in machine learning and agricultural weed
3. Implementation and Testing: The researcher will implement the planned features, following industry coding standards and performing thorough testing to guarantee the system’s functionality and
4. Review and Adaptation: Following each iteration, the researcher will assess the progress and make necessary adjustments to the system using insights from educational resources written by agricultural experts. This iterative process guarantees that the system stays in sync with domain requirements and integrates the most recent advancements in the

During the development process, the researcher will prioritise continuous learning and adaptation. They will actively search for educational materials, like videos and articles written by agricultural experts, to improve their understanding of the field and ensure the system’s accuracy and relevance. Thanks to its iterative and adaptive nature, the Agile methodology allows researchers to seamlessly integrate domain knowledge from educational resources into the development process. This approach will lead to the development of an intelligent weed detection system that is both technically strong and well-suited to the practical needs and difficulties of the agricultural field.

System Requirements

Requirement Elicitation and Specifications

The researcher will systematically gather and define the requirements for the Intelligent Weed Detection Using Machine Learning system. As part of this procedure, input will be collected from various stakeholders, including agricultural experts, farmers, and computer vision and machine learning researchers.

Requirement Elicitation

The requirement elicitation process will involve the following steps:

1. Stakeholder Identification: The researcher aims to identify the primary stakeholders impacted by or with a vested interest in the intelligent weed detection system’s development. In this context, the stakeholders typically consist of agricultural experts, farmers, agricultural technology companies, and researchers in related
2. Domain Knowledge Acquisition: The researcher will gather domain knowledge from educational resources, including articles, books, and videos created by agricultural Understanding the complexities of weed detection, crop growth stages, and the potential impact of the proposed system on agricultural practices is essential knowledge.
3. Requirements Analysis: The collected data and domain knowledge will be analysed to determine and prioritise the system’s functional and non-functional requirements. This analysis will examine various factors, including accuracy, performance, usability, scalability, and environmental

Requirements Specification

Based on the elicited requirements, the researcher will document an in-depth set of specifications for the Intelligent Weed Detection Using Machine Learning system. The requirements specification will consist of the following components:

1.   Functional Requirements: These requirements will outline the precise capabilities and functionalities that the system must have. Examples of functional requirements may include:

• Accurately detecting and classifying various weed species and crop types
• Handling diverse environmental conditions, such as lighting variations and overlapping plants
• Integrating with existing agricultural machinery and systems
• Providing real-time or near real-time weed detection capabilities

2.   Non-Functional Requirements: These requirements will specify the system’s qualitative attributes and constraints. Examples of non-functional requirements may include:

• Performance and scalability requirements to handle large-scale agricultural operations
• Usability and user interface design considerations for ease of use by farmers and agricultural workers
• Reliability and fault tolerance to ensure consistent and accurate operation
• Environmental impact considerations, such as minimising the use of herbicides and promoting sustainable practices

3.   System Constraints: This section will outline the constraints that the system must be subject to, including limitations on computational resources, data availability, and regulatory or legal obligations.

4. System Interfaces: The specification will establish the interfaces by which the system will engage with other components, including data sources, agricultural machines, and user interfaces.

The requirements specification will act as a thorough guide for the development, implementation, and testing of the Intelligent Weed Detection Using Machine Learning system. Its purpose is to ensure that the system meets the needs and expectations of all stakeholders and promotes sustainable and environmentally friendly agricultural practices.

REQUIREMENTS MODEL

The researcher will utilize a structured requirements modelling approach to effectively represent and organize the elicited requirements for the Intelligent Weed Detection Using Machine Learning system. The requirements model presents a clear and organized view of the system’s functionalities, constraints, and dependencies. This promotes effective communication and understanding among stakeholders.

The researcher will employ the Unified Modelling Language (UML) and its corresponding diagramming techniques to construct the requirements model. The UML diagrams that will be utilised are as follows: Activity diagrams will model the system’s workflow and decision-making processes. These diagrams will be handy in representing the logic and control flows involved in tasks like image analysis, weed classification, and decision-making for weed management strategies. UML diagrams can be used to develop a thorough requirements model. This model will clearly and succinctly represent the system’s requirements, enabling effective communication with stakeholders and ensuring a shared understanding of the system’s functionalities, constraints, and architectural components. The requirements model will provide a solid basis for the later stages of system design, implementation, and testing. This will allow for creating an Intelligent Weed Detection Using a Machine Learning system that effectively addresses stakeholders’ requirements and encourages sustainable and efficient agricultural practices.

DISCUSSION AND CONCLUSION

The researcher employed a systematic and comprehensive analysis approach to identify potential issues and challenges that may arise during the development of the Intelligent Weed Detection Using Machine Learning system. This approach was designed to minimize risks and ensure a seamless implementation process. A significant challenge identified is the difficulty in accurately differentiating between crops and weeds, especially during their initial growth stages. During this phase, plants display notable visual similarities, which poses a challenge for traditional image processing methods to distinguish them effectively. The researcher acknowledged the necessity for advanced machine learning algorithms and deep learning architectures that can effectively extract and interpret subtle features from image data. To tackle this challenge, the researcher thoroughly reviewed the existing literature to investigate the latest techniques in weed detection and machine learning.

The review emphasized the effectiveness of convolutional neural networks (CNNs) and transfer learning approaches, which have shown impressive results in image classification tasks. Furthermore, the review highlighted the significance of carefully selecting various crop and weed species, environmental conditions, and growth stages to ensure a comprehensive and inclusive dataset. One notable challenge identified is the possibility of biased or incomplete datasets. This could result in less-than-optimal model performance and inaccuracies in weed detection. The researcher acknowledged the importance of a diligent data collection and annotation process to guarantee the quality and diversity of the dataset. To address this challenge, the researcher devised a comprehensive data-gathering method that included acquiring images from various sources, such as publicly accessible repositories and collaborations with agricultural specialists. The researcher implemented a thorough annotation and labelling process, using educational resources and domain knowledge from agricultural specialists to precisely identify and document relevant information, including plant species, growth stages, and environmental conditions. The researcher also noted the possibility of overfitting, a common issue in machine learning models. Overfitting occurs when the algorithm becomes overly focused on the training data and struggles to apply its learnings to new, unseen data.

To reduce this risk, the researcher intended to use data augmentation techniques, including rotation, flipping, and color jittering, to enhance the training data’s diversity and the model’s generalization. In addition, the researcher acknowledged the importance of implementing an in-depth assessment and testing approach to guarantee the system’s dependability and precision in real-world scenarios. The Ishikawa (Fishbone) diagram, shown in Figureure 12, was used to identify the underlying reasons for ineffective and environmentally damaging weed control methods. The diagram identified key factors contributing to the main problem, including labor-intensive manual weeding, soil contamination, difficulty distinguishing crops from weeds, and high input costs. After careful analysis, the researcher devised a detailed plan to tackle these challenges. This plan includes a hybrid research methodology, an agile system development approach, and a clear requirements model. The hybrid research methodology used a combination of experimental and quantitative approaches to develop and evaluate machine learning models. It also incorporated statistical analysis and data-driven decision-making to ensure a comprehensive and iterative process. The agile system development methodology allowed for iterative and incremental development, allowing the researcher to adapt to changing requirements and incorporate stakeholder feedback. The requirements model, created with the Unified Modelling Language (UML), offered a precise and structured representation of the system’s functionalities, constraints, and dependencies. This facilitated efficient communication and understanding among stakeholders. The researcher effectively applied a rigorous analysis method and proactively anticipated potential challenges, enabling them to successfully address the complexities of developing an intelligent weed detection system. The analysis provided a solid basis for the following stages of the project, guaranteeing a systematic and knowledgeable approach to tackling the challenges and reaching the desired objectives.

REFERENCES

1. Adeniji, A. et al. (2023) ‘Deployment of an artificial intelligent robot for weed management in legumes farmland’, ABUAD Journal of Engineering Research and Development, 6(2), pp. 28–38.
2. Ahmed, et al. (2012) ‘Classification of crops and weeds from digital images: A support vector machine approach’, Crop Protection, 40, pp. 98–104.
3. Arakeri, M.P. et al. (2017) ‘Computer vision based robotic weed control system for precision agriculture’, in 2017 International Conference on Advances in Computing, Communications and Informatics (ICACCI). IEEE, 1201–1205.
4. Bah, M.D., Hafiane, and Canals, R. (2018) ‘Deep learning with unsupervised data labeling for weed detection in line crops in UAV images’, Remote sensing, 10(11), p. 1690.
5. Cicco, M. et al. (2017) ‘Automatic model based dataset generation for fast and accurate crop and weeds detection’, in 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 5188–5195.
6. Coleman, G.R.Y. et al. (2022) ‘Weed detection to weed recognition: reviewing 50 years of research to identify constraints and opportunities for large-scale cropping systems’, Weed Technology, 36(6), pp. 741–757.
7. Esposito, M. et al. (2021) ‘Drone and sensor technology for sustainable weed management: A review’, Chemical and Biological Technologies in Agriculture, 8, 1–11.
8. Etienne, A. and Saraswat, D. (2019) ‘Machine learning approaches to automate weed detection by UAV based sensors’, in Autonomous air and ground sensing systems for agricultural optimization and phenotyping SPIE, pp. 202–215.
9. Firmansyah, E. et al. (2022) ‘Real-time Weed Identification Using Machine Learning and Image Processing in Oil Palm Plantations’, in IOP Conference Series: Earth and Environmental Science. IOP Publishing, 012046.
10. Gallo, I. et al. (2023) ‘Deep object detection of crop weeds: Performance of YOLOv7 on a real case dataset from UAV images’, Remote Sensing, 15(2), 539.
11. Gašparović, M. et al. (2020) ‘An automatic method for weed mapping in oat fields based on UAV imagery’, Computers and electronics in agriculture, 173, 105385.
12. Ibrahim, et al. (2022) ‘New hybrid invasive weed optimization and machine learning approach for fault detection’, Energies, 15(4), p. 1488.
13. Islam, N. et al. (2021) ‘Early weed detection using image processing and machine learning techniques in an Australian chilli farm’, Agriculture, 11(5), p.
14. Junior, L.C.M. and Ulson, J.A.C. (2021) ‘Real time weed detection using computer vision and deep learning’, in 2021 14th IEEE International Conference on Industry Applications (INDUSCON). IEEE, 1131–1137.
15. Khan, S. et (2021) ‘Deep learning-based identification system of weeds and crops in strawberry and pea fields for a precision agriculture sprayer’, Precision Agriculture, 22(6), pp. 1711–1727.
16. i, N. et al. (2019) ‘Review of machine-vision-based plant detection technologies for robotic weeding’, in 2019 IEEE International Conference on Robotics and Biomimetics (ROBIO). IEEE, 2370–2377.
17. Liakos, G. et al. (2018) ‘Machine learning in agriculture: A review’, Sensors, 18(8), p. 2674.
18. Lin, F. et al. (2017) ‘Detection of corn and weed species by the combination of spectral, shape and textural features’, Sustainability, 9(8), 1335.
19. Liu, and Bruch, R. (2020) ‘Weed detection for selective spraying: a review’, Current Robotics Reports, 1, pp. 19–26.
20. Liu, Y. (2022) ‘(Retracted) Field weed recognition algorithm based on machine learning’, Journal of Electronic Imaging, 31(5), 51413.
21. Lottes, P. et al. (2020) ‘Robust joint stem detection and crop‐weed classification using image sequences for plant‐specific treatment in precision farming’, Journal of Field Robotics, 37(1), pp. 20–34.
22. Mohidem, A. et al. (2021) ‘How can unmanned aerial vehicles be used for detecting weeds in agricultural fields?’, Agriculture, 11(10), p. 1004.
23. Murad, N.Y. et al. (2023) ‘Weed detection using deep learning: a systematic literature review’, Sensors, 23(7), 3670.
24. Olsen, A. (2020) ‘Improving the accuracy of weed species detection for robotic weed control in complex real-time environments’, in. Available at: https://api.semanticscholar.org/CorpusID:235048677.
25. Pantazi, X.-E., Moshou, D. and Bravo, C. (2016) ‘Active learning system for weed species recognition based on hyperspectral sensing’, Biosystems Engineering, 146, 193–202.
26. Rani, S.V.J. et al. (2022) ‘Automated weed detection system in smart farming for developing sustainable agriculture’, International Journal of Environmental Science and Technology, 19(9), pp. 9083–9094.
27. Sarvini, T. et al. (2019) ‘Performance comparison of weed detection algorithms’, in 2019 International Conference on Communication and Signal Processing (ICCSP). IEEE, 843–847.
28. Shaw, D.R. (2005) ‘Remote sensing and site‐specific weed management’, Frontiers in Ecology and the Environment, 3(10), pp. 526–532.
29. Steininger, D. et al. (2023) ‘The cropandweed dataset: A multi-modal learning approach for efficient crop and weed manipulation’, in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 3729–3738.
30. Li, N. et al. (2019) ‘Review of machine-vision-based plant detection technologies for robotic weeding’, in 2019 IEEE International Conference on Robotics and Biomimetics (ROBIO). IEEE, 2370–2377.
31. Liakos, G. et al. (2018) ‘Machine learning in agriculture: A review’, Sensors, 18(8), p. 2674.
32. Lin, F. et al. (2017) ‘Detection of corn and weed species by the combination of spectral, shape and textural features’, Sustainability, 9(8), 1335.
33. Liu, and Bruch, R. (2020) ‘Weed detection for selective spraying: a review’, Current Robotics Reports, 1, pp. 19–26.
34. Liu, Y. (2022) ‘(Retracted) Field weed recognition algorithm based on machine learning’, Journal of Electronic Imaging, 31(5), 51413.
35. Lottes, P. et al. (2020) ‘Robust joint stem detection and crop‐weed classification using image sequences for plant‐specific treatment in precision farming’, Journal of Field Robotics, 37(1), pp. 20–34.
36. [35] Mohidem, A. et al. (2021) ‘How can unmanned aerial vehicles be used for detecting weeds in agricultural fields?’, Agriculture, 11(10), p. 1004.
37. Murad, N.Y. et al. (2023) ‘Weed detection using deep learning: a systematic literature review’, Sensors, 23(7), 3670.
38. Olsen, A. (2020) ‘Improving the accuracy of weed species detection for robotic weed control in complex real-time environments’, in. Available at: https://api.semanticscholar.org/CorpusID:235048677.
39. Pantazi, X.-E., Moshou, D. and Bravo, C. (2016) ‘Active learning system for weed species recognition based on hyperspectral sensing’, Biosystems Engineering, 146, 193–202.
40. Rani, S.V.J. et al. (2022) ‘Automated weed detection system in smart farming for developing sustainable agriculture’, International Journal of Environmental Science and Technology, 19(9), pp. 9083–9094.
41. Sarvini, T. et al. (2019) ‘Performance comparison of weed detection algorithms’, in 2019 International Conference on Communication and Signal Processing (ICCSP). IEEE, 843–847.
42. Shaw, D.R. (2005) ‘Remote sensing and site‐specific weed management’, Frontiers in Ecology and the Environment, 3(10), pp. 526–532.
43. Steininger, D. et al. (2023) ‘The cropandweed dataset: A multi-modal learning approach for efficient crop and weed manipulation’, in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 3729–3738.
44. Su, W.-H. (2020) ‘Advanced machine learning in point spectroscopy, RGB-and hyperspectral- imaging for automatic discriminations of crops and weeds: A review’, Smart Cities, 3(3), pp. 767–
45. Ukaegbu, U.F. et (2021) ‘Deep learning application in diverse fields with plant weed detection as a case study’, in Proceedings of the International Conference on Artificial Intelligence and Its Applications, pp. 1–9.
46. Umamaheswari, S., Arjun, R. and Meganathan, D. (2018) ‘Weed detection in farm crops using parallel image processing’, in 2018 Conference on Information and Communication Technology (CICT). IEEE, 1–4.
47. Urmashev, B. et al. (2021) ‘Development of a weed detection system using machine learning and neural network algorithms’, Eastern-European Journal of Enterprise Technologies, 6(2), 114.
48. Wang, A., Zhang, W. and Wei, X. (2019) ‘A review on weed detection using ground-based machine vision and image processing techniques’, Computers and electronics in agriculture, 158, 226–240.
49. Wang, Y. (2023) ‘A Deep Learning-based Approach for Vision-based Weeds Detection’, International Journal of Advanced Computer Science and Applications [Preprint]. Available at: https://api.semanticscholar.org/CorpusID:266647522.
50. Van Der Weide, Ry. et al. (2008) ‘Innovation in mechanical weed control in crop rows’, Weed research, 48(3), 215–224.
51. Wu, X. et al. (2020) ‘Robotic weed control using automated weed and crop classification’, Journal of Field Robotics, 37(2), pp. 322–340.
52. Wu, Z. et al. (2021) ‘Review of weed detection methods based on computer vision’, Sensors, 21(11), 3647.
53. Yang, Y. et al. (2022) ‘Dissimilarity-based active learning for embedded weed identification’, Turkish Journal of Agriculture and Forestry, 46(3), 390–401.