Department of Computer Science and Engineering

JP Nagar, Kothanur, Bengaluru - 560076

GRASS WEED DETECTION

Academic year 2024 – 2025 Odd (7th) semester

Major Project (21CSP76)
Review Presentation

Presented by: Under the Guidance of:

Aryan P Naik (1RF21CS018) Prof. Padmasree N
Daivik Simha K N (1RF21CS028) Asst. Professor
Dhruv Vijay Vittal (1RF21CS037) Dept. of CSE
Harshith T (1RF21CS048) RVITM
 CONTENTS

❖ Introduction
❖ Literature Survey
❖ Methodology
❖ Ensemble Methods for Weed Detection
❖ Heatmap and Overlay
❖ Advantages
❖ Future Outcome
❖ References
 INTRODUCTION
Why Grass Weed Detection Matters?
• Threat to Crop Growth: Grass weeds compete with crops for essential resources like water, nutrients, and sunlight, leading
to reduced yields.
• Impact on Productivity: Uncontrolled weed growth can cause significant economic losses for farmers.

Challenges in Traditional Weed Detection

• Labor-Intensive: Manual weed detection requires significant human effort and time.
• Time-Consuming: Large-scale farms make manual weed removal impractical.
• Inefficiency: Traditional methods often lead to inaccurate detection and delayed action, reducing crop health.

Why Automation is Needed?

• Faster and More Efficient: Automated systems can quickly scan large fields with high accuracy.
• Cost-Effective: Reduces dependency on manual labor and frequent herbicide applications.
• Improved Precision: AI-based detection ensures better weed classification, minimizing crop damage.
Project Goal

• Develop an automated grass weed detection system to assist farmers in efficient weed management.

• Enhance agricultural productivity by reducing manual efforts and improving accuracy.

Technology Used

• Convolutional Neural Networks (CNNs): A deep learning model that excels at image recognition and classification.

• Image Processing Techniques: Enhances detection accuracy by filtering and analyzing field images.

• Machine Learning Algorithms: Helps in continuously improving detection efficiency over time.

Training Data & Functionality

• Diverse Dataset: System is trained on real agricultural field images containing different types of weeds and crops.

• Accurate Classification: Distinguishes grass weeds from crops and non-weed vegetation.

• Scalable Solution: Can be integrated with drones, robots, or farm machinery for large-scale deployment.
 Literature Survey
Year Title Author(s) Description Advantages Drawbacks
2023 Review on Weed Nafeesa his paper presents a systematic literature Standardized review process with concrete Limited to articles published between 2015 and
Detection Yousuf review (SLR) on weed detection techniques research questions. 2023.
Techniques in Murad, Tariq in smart farming using deep learning. The Comprehensive analysis of different Heavy reliance on deep learning methods,
Smart Farming Mahmood, review focuses on the trends, types of algorithmic performances. potentially overlooking other effective
Using Deep Ahsan weeds and crops detected, and the Detailed guideline for current performance and techniques.
Learning[1] Morshed performance of various deep learning future research directions.
algorithms.

2021 Weed Detection Atshaya G, This document discusses various smart Eliminates the need for manual crop scouting. Does not provide specific experimental results
Using Image Hemalatha M farming techniques for weed detection using Utilizes advanced DL models for higher or data.
Processing machine learning, particularly deep learning accuracy in weed detection. Lacks detailed information on the authors and
[2] (DL). publication year.

2022 Weed detection by Vijaykumar This paper reviews various image Comprehensive review of CNN-based weed Limited to CNN and image processing
using image Bidve, processing techniques for weed detection, detection methods. techniques.
processing [3] Sulakshana emphasizing convolutional neural networks Discusses integration of drone technology and The summary is brief and lacks detailed
Mane, Pradip (CNNs). It provides a summary of recent deep learning. experimental results.
Tamkhade literature, highlighting different methods.

2023 Detection of Weeds P. Kavitha This research presents an image processing High accuracy in detecting multiple types of Potentially high computational complexity and
by Using Machine Reddy, R. technique utilizing deep convolutional weeds. resource requirements.
Learning[4] Anirudh neural networks (CNN) to detect weeds in Reduction in herbicide use, contributing to Requires high-quality images and extensive
Reddy, agricultural fields. Images of various crops environmental sustainability. training data for optimal performance.
Katkam Sai are captured and analyzed to classify
Teja, K. affected regions.
Rohith, and K.
Rahul
Year Title Author(s) Description Advantages Drawbacks
2024 A two-stage deep learning Felipe Calder This research presents a two-stage weed This two-stage weed identification strategy A drawback of this two-stage weed identification
strategy for weed ara- identification system using visible spectrum offers high accuracy (97.16%) with minimal approach is that it focuses primarily on detecting
identification in Cea, Miguel T images. In the first stage, a convolutional data, making it efficient for real-time use. Its specific weed types, which may limit its
grassfields orres- Siamese neural network identifies potential weed lightweight model reduces the need for large generalizability to other species or environments.
Torriti, Ferna regions. The second stage uses a convolutional datasets, enabling deployment on embedded Additionally, while it performs well with a small
ndo Auat neural network to confirm irregular weeds. The systems. Additionally, it performs well in dataset, further testing is needed to assess its
Cheein , José approach achieves a 97.16% accuracy and an varied conditions, enhancing practicality for scalability with larger datasets or more diverse
Delpiano 81.14% macro F1 score. agricultural applications. conditions.

2024 A Comprehensive Review *Prof. The agricultural industry requires sustainable The advantages of image-based weed detection Image-based weed detection with deep learning
of Weed Detection Sowmya.1, weed management techniques beyond using deep learning include enhanced has notable drawbacks. It depends on high-
through Advanced Image Dr. Sandeep conventional pesticides, which can harm crops precision in targeting weeds without harming quality labeled datasets that are difficult to
Processing and Deep Bhat2 and the environment. This paper reviews crops, improved efficiency over traditional obtain, and it requires significant computational
Learning advancements in image-based weed detection methods, and adaptability to different resources that may not be accessible to all
using deep learning algorithms. agricultural conditions. farmers.

2023 Designing and Hossein This study aims to design an object detection Automates weed management, saving time and Requires high-quality datasets and extensive
Developing a Weed Chegini, model for weed monitoring and control in costs for farmers. hyperparameter tuning.
Detection Model for Fernando pastures, focusing on California thistle. A High accuracy in detecting California thistle. Potential initial setup costs and technical
California Thistle[19] Beltran, synthetic methodology created plant-based, leaf- challenges in implementation.
Aniket based, and mixed datasets. The leaf-based model
Mahanti achieved 93% mAP accuracy on training images
and over 95% on testing images.

2019 Performance Comparison Sarvini T, The proposed system automates weed detection Reduces the need for manual labor in weed
of Weed Detection Sneha T, in four commercial crops using machine detection.
Requires a large, high-quality dataset for
Algorithms[20] Sukanya learning. It employs image processing Improves accuracy and efficiency in
Gowthami techniques, including the Excess Green method identifying weeds. effective training.
G.S, and Otsu’s thresholding, to isolate weeds like Leverages advanced machine learning Potentially high computational demands and
Sushmitha S Para grass and Nutsedge. techniques for better performance. complexity in implementation.
Year Title Author(s) Description Advantages Drawbacks
2021 Cascading Feature
a
A. This study compares different classifiers for Improved classification accuracy using The document does not specify particular
Filtering and Boosting Bakhshipour plant type classification using image features. It boosted models. limitations or challenges in the provided extract.
Algorithm for Plant highlights the superior performance of the Effective integration into computer vision
Type Classification LogitBoost-IG-RF model. The study suggests systems for early weed detection.
Based on Image integrating this boosted classifier into computer Provides a comparative analysis of different
Features[13] vision systems for enhanced weed detection. classifiers.

2024 Efficient Application of Harish Kumar This paper explores the use of deep neural High precision in detecting small and multiple Specific drawbacks are not mentioned in the
Deep Neural Networks networks to identify small and multiple weed weed patches. provided text, but typical challenges may include
for Identifying Small patches from drone images. The methodology Utilizes drone images for large-scale computational complexity and the need for high-
and Multiple Weed improves the precision of weed detection in agricultural monitoring. quality image data.
Patches Using Drone agricultural settings. The study emphasizes the Enhances weed management practices.
Images benefits of drone-based imaging for scalable
[14] weed management.

2023 Saliency-Based Muhammad This study utilizes UAV multispectral imaging

Semantic Weeds javaid aslam, for the detection and classification of weeds. The Utilizes advanced multispectral imaging. High costs associated with UAV and
Detection and Sajjad saliency-based approach aims to improve the Improves the accuracy of weed detection and multispectral imaging equipment.
Classification Using manzoor, accuracy of weed identification. The method classification. Potential challenges in processing and analyzing
UAV Multispectral Tasleem leverages advanced imaging technologies to Applicable for large-scale agricultural large volumes of imaging data.
Imaging[15] kausar enhance agricultural monitoring. monitoring. Dependence on favorable weather conditions for
UAV operations.

2021 Weed Identification Xiaojun Jin, The study focuses on the application of deep Effective weed identification using deep Requires extensive computational resources and
Using Deep Learning Jun Che, learning and image processing for weed learning. high-quality datasets.
and Image Processing Yong Chen. identification in vegetable plantations. It Enhances precision in distinguishing weeds Potential challenges with varying lighting and
in Vegetable highlights the effectiveness of these techniques from crops. environmental conditions.
Plantation[16] in distinguishing weeds from crops. The research Supports improved agricultural practices. Implementation complexity may require
aims to enhance precision agriculture practices. specialized knowledge and equipment.
Year Title Author(s) Description Advantages Drawbacks
2025 Harnessing Artificial G. S. The paper discusses the integration of artificial - Broad perspective on AI applications in - Lacks specific implementation details
Intelligence and Remote Mmbando intelligence and remote sensing technologies in agriculture - May be more theoretical than practical
Sensing in Climate- climate-smart agriculture, emphasizing strategies - Highlights the role of remote sensing in weed - Limited focus on weed detection methodologies
Smart Agriculture: The to enhance global food security through detection
Current Strategies improved weed management. - Addresses global food security concerns
Needed for Enhancing
Global Food Security[9]
2024 A Lightweight Model J. Huang, X. This study introduces an improved YOLOv8s - High detection accuracy - May require extensive training data
for Weed Detection Xia, Z. Diao, network tailored for weed detection in maize - Lightweight model suitable for real-time - Performance may decline in different crop
Based on the Improved X. Li, S. fields, aiming for a balance between accuracy applications environments
YOLOv8s Network in Zhao, J. and computational efficiency. - Specifically optimized for maize fields - Potential challenges in detecting overlapping
Maize Fields[10] Zhang, B. weeds
Zhang, G. Li

2024 A Comprehensive F. D. This survey provides an extensive overview of - Extensive coverage of recent research - May lack practical implementation details
Survey on Weed and
Crop Classification
a
Adhinata,
Wahyono, R.
machine learning and deep learning techniques
applied to weed and crop classification,
- Highlights both machine learning and deep
learning approaches
- Could be overwhelming due to breadth
- Rapid field advancements may outdate some
Using Machine Sumiharto highlighting recent advancements and - Discusses challenges and future directions information
Learning and Deep challenges.
Learning[11]

2019 Automatic Detection of Elena C. his study investigates the automatic detection of Enhances accuracy in weed detection. Enhances accuracy in weed detection.
Aquatic Weeds: A Case Rodríguez- aquatic weeds in the Guadiana River in Spain. Provides a method for large-scale Provides a method for large-scale environmental
Study in the Guadiana Garlito,Abel The aim is to develop a method for efficient and environmental monitoring. monitoring.
River.[12] Paz- accurate identification of invasive weed species Contributes to better management strategies Contributes to better management strategies for
Gallardo , and using advanced imaging and analysis techniques. for invasive species. invasive species.
Antonio Plaza The findings suggest improvements in
environmental monitoring and management
strategies.
Year Title Author(s Description Advantages Drawbacks
)
2021 Review of Weed Zhangnan This review explores various methods used in Provides a comprehensive overview of weed Traditional image-processing methods require
Detection Methods Wu, Yajun weed detection through computer vision, detection methods, including both traditional manual feature design and are highly dependent
Based on Computer Chen, Bo focusing on both traditional image-processing and deep learning approaches. on image acquisition quality.
Vision[5] Zhao, techniques and modern deep learning Discusses the use of public image datasets for Deep learning methods need large datasets for
Xiaobing approaches. It highlights the advantages and precision agriculture tasks, promoting the training, making image collection challenging
Kang disadvantages of these methods. development of computer vision technology in and labor-intensive.
agriculture.
2023 Machine Learning and H. Pathak, The study compares machine learning - Comparative analysis provides insights into - May not generalize to other crops
Handcrafted Image C. algorithms with handcrafted image processing method effectiveness - Handcrafted methods may lack adaptability
Processing Methods for Igathinathan techniques for classifying common weeds in - Addresses common weeds in corn fields - Machine learning models require substantial
Classifying Common e, K. corn fields. - Combines modern and traditional approaches training data
Weeds in Corn Field[6] Howatt, Z.
Zhang

2022 Automatic Detection of E. C. The paper presents a study on automatic - Addresses aquatic weed detection - Specific to aquatic environments
Aquatic Weeds: A Case Rodríguez- detection of aquatic weeds using remote sensing - Utilizes remote sensing data - May not generalize to terrestrial weed detection
Study in the Guadiana Garlito, A. data, focusing on the Guadiana River in Spain. - Case study provides practical insights - Dependent on quality of remote sensing data
River, Spain Paz-
Gallardo, A.
Plaza

2021 Review of Weed Zhangnan This review article examines various computer Comprehensive review of both traditional and The review might be less detailed on specific
Detection Methods Wu, Yajun vision methods for weed detection, including deep learning-based weed detection methods. methodologies due to the broad scope.
Based on Computer Chen, Bo traditional image-processing techniques and Discusses a wide range of datasets and tools Limited practical application insights as it is
Vision[8] Zhao, deep learning-based approaches. It discusses the available for weed detection research. more theoretical and review-based.
Xiaobing advantages and disadvantages of these methods,
Kang, analyzes current challenges in weed detection.
 METHODOLOGY
To achieve the objective of developing an efficient and accurate machine learning model for automated weed detection, we will follow a structured
methodology. This involves leveraging a comprehensive agricultural dataset and utilizing advanced classifier algorithms. The methodology
encompasses several critical steps, including data collection, preprocessing, model selection, training, evaluation, and deployment. Each step is
designed to ensure the creation of a robust model capable of accurately identifying weeds, thereby enhancing crop management efficiency and
supporting sustainable agriculture.

Dataset Collection

Utilize an agricultural dataset: Gather images representing various types of weeds and non-weeds from a reliable agricultural dataset.

Data Preprocessing

• Clean the dataset: Ensure the dataset is free from duplicates and irrelevant images.
• Resize images: Resize all images to 139x139 pixels to match the input size of the InceptionV3 model.
• Normalize pixel values: Scale pixel values to the range [0, 1] by dividing by 255.
• Data augmentation: Apply techniques such as rotation, width and height shifts, shear transformations, zooming, and horizontal flipping to
enhance model generalization.
Data Annotation

• Accurate labelling: Ensure accurate labelling of images as "Weed" or "Not Weed".

• Split the data: Divide the dataset into training (70%), validation (20%), and test sets (10%).

Model Selection

• Base Model: Utilize the pre-trained InceptionV3 and ResNet architecture without the top layers.
• Custom Layers: Add custom dense layers on top of InceptionV3 to adapt it for binary classification.

Model Training

• Compile the model: Use the Adam and RMSprop optimizer with a learning rate of 0.001 and categorical cross-entropy loss.
• Training and validation: Train the model on the training set and validate it on the validation set using a batch size of 128 and for 30
epochs.
• Callbacks: Implement callbacks such as early stopping, learning rate scheduler, model checkpoint, and reduce learning rate on plateau
to enhance training efficiency and prevent overfitting.
 Model Evaluation

• Assess performance: Evaluate the model's performance on the test set using metrics such as accuracy, precision, recall, and F1-score.

• Conduct robustness testing under varying environmental conditions and weed appearances.

Web Application Development

 Develop a user-friendly web application using the Flask framework. Provide features for farmers to upload images and receive real-time
detection results.

 Include heatmaps, overlays, and tools for adjusting detection parameters and exploring detection scenarios.

Deployment and Testing

• Deploy the ensemble model and web application in real-world scenarios. Gather feedback from users and stakeholders to evaluate accuracy,
robustness, and usability. Refine and improve the system based on feedback, addressing issues and exploring further enhancements.
Fig. Methodology Flowchart
 ENSEMBLE METHODS FOR WEED DETECTION
Ensemble methods combine predictions from multiple models to improve accuracy and robustness. In weed detection, they help in creating more
reliable predictions by reducing errors that a single model might make.
How It Works:
Different models, such as Convolutional Neural Networks (CNNs), Decision Trees, or Support Vector Machines, are trained individually. Each
model generates its own prediction, and these predictions are combined (e.g., via averaging, majority voting, or weighted methods) to produce a
final result.
STEP 1: MODEL SELECTION FOR ENSEMBLING

InceptionV3 (Adam Optimizer)

Why Chosen?

• Designed for efficient feature extraction and handling complex patterns in images.

• Well-suited for detecting small and intricate weed structures within agricultural fields.

Adam Optimizer Advantage:

• Adaptive learning rate enhances model convergence speed and stability.

• Effectively balances precision and recall for improved classification.

InceptionV3 (RMSprop Optimizer)
• Shares the same powerful architecture as InceptionV3 (Adam) but optimized differently.

• Helps in better capturing high-frequency features in image data.

RMSprop Optimizer Advantage:

• Handles noisy and complex data efficiently.

• Improves weight updates for a smoother training process.

EfficientNetB0 (Adam Optimizer)

• Designed for optimized performance, achieving high accuracy with fewer parameters.

• Provides better scalability, making it ideal for large-scale weed detection.

Adam Optimizer Advantage:

• Reduces overfitting while maintaining high classification accuracy.

• Ensures a fast and stable learning process across diverse datasets.

 Voting System: Majority-Based Decision Making

To enhance the accuracy and reliability of grass weed detection, an ensemble learning approach is implemented using a majority voting
mechanism. This ensures robust predictions by leveraging multiple models instead of relying on a single one.

How the Voting System Works?

1. Independent Predictions:
• Each of the three pre-trained CNN models (InceptionV3 (Adam), InceptionV3 (RMSprop), and EfficientNetB0 (Adam)) independently
analyzes the input image and predicts its class (Weed or Not Weed).
2. Majority Vote Decision:
• The final classification is determined based on the majority vote among the three models.
• Example Scenarios:
• Case 1: If two models predict "weed" and one predicts "crop," the final decision is "weed."
• Case 2: If all three models predict the same class, that class is chosen with maximum confidence.
Why Use Majority Voting?
 Minimizes Individual Model Errors:
If one model makes an incorrect prediction due to noise or poor lighting, the majority vote helps override that mistake.
 Increases Prediction Stability:
By combining multiple models, the system avoids relying on the weaknesses of a single architecture.
 Enhances Generalization:
Different models extract diverse features, leading to better performance across varied weed appearances and
environmental conditions.
Why Use an Ensemble Approach?

🔹 Combining Diverse Strengths: Different CNN architectures (InceptionV3 & EfficientNetB0) specialize in varied feature
extraction. InceptionV3 excels at recognizing complex patterns, while EfficientNetB0 optimizes detection with fewer parameters. This multi-
model approach ensures better weed classification across diverse environmental conditions.

🔹 Improved Generalization & Robustness: Captures a wider range of patterns in weed structures, soil textures, and lighting
variations. Works effectively across different field conditions, weed species, and crop types. Performance Enhancement & Accuracy Boost

• Higher Precision & Recall: Maintains a balanced trade-off to reduce false positives and false negatives.

• Resilience to Noise & Anomalies: Handles blurry images, occlusions, and variable backgrounds effectively.

• Reliable & Scalable: Adaptable for large-scale farming with minimal human intervention.
 HEATMAP AND OVERLAY
Heatmaps and overlays enhance visualization, making model outputs intuitive for users. They highlight weed-affected areas, aiding decision-
making and resource allocation. By providing clear detection insights, these tools improve weed management efficiency and bridge the gap
between technical results and practical application.

Heatmap in Weed Detection

Heatmaps visualize the model’s confidence in weed detection using color gradients. Warmer colors (red, orange) indicate high weed presence,
while cooler tones (blue) suggest lower confidence. This intuitive representation helps farmers quickly identify affected areas. Adjustable
confidence thresholds provide flexibility in decision-making, and overlaying heatmaps on the original image enhances contextual understanding
of field conditions.

Overlay in Weed Detection

Overlays directly map the model’s predictions onto the original field image, visually marking weed-affected areas for easy identification. These
highlighted regions, outlined or shaded, help farmers pinpoint problem spots, especially in large fields. By providing clear visual feedback,
overlays simplify interpretation and enable quick, targeted interventions, ensuring efficient resource allocation for weed management.
 ADVANTAGES

• Robustness to Noisy Data: Handles variations in lighting, occlusions, and image distortions effectively.

• Error Reduction: Mitigates the impact of individual model weaknesses by leveraging diverse architectures.

• Scalability: Adaptable for large-scale agricultural applications with minimal performance degradation.

• Adaptability to Different Conditions: Works efficiently across varying soil textures, crop types, and weed growth stages.

• Improved Confidence in Predictions: The majority voting mechanism ensures more consistent and reliable classification.

• Optimized Trade-off Between Speed & Accuracy: Balances model complexity while maintaining high detection precision.
 FUTURE OUTCOME

High Accuracy in Weed Detection: Achieve exceptional accuracy in identifying and classifying various types of grass weeds, significantly
reducing misclassification rates compared to manual methods.

Enhanced Crop Management Efficiency: Streamline the weed management process, enabling precise targeting of weed-infested areas and
optimizing resource usage.

Manual Labor Reduction: Substantially decrease the need for manual weed identification, allowing farmers to focus on other critical-tasks.

Promotion of Sustainable Farming Practices: Reduce the reliance on chemical herbicides by implementing targeted weed control, contributing
to environmental conservation and sustainable agriculture.

Increased Crop Yields: Improve crop productivity by effectively managing weed competition, leading to higher yields.

Data-Driven Agricultural Decisions: Provide actionable insights and valuable data for informed decision-making on weed control strategies and
resource allocation, enhancing overall farm management.
 REFERENCES
[1] N. Y. Murad, T. Mahmood, A. R. M. Forkan, A. Morshed, P. P. Jayaraman, and M. S. Siddiqui, "Weed detection using deep learning: A systematic literature
review," Sensors, vol. 23, no. 7, p. 3670, 2023, doi: [10.3390/s23073670](https://doi.org/10.3390/s23073670).

[2] H. M. Sahin, T. Miftahushudur, B. Grieve, and H. Yin, "Segmentation of weeds and crops using multispectral imaging and CRF-enhanced U-Net,"
Computers and Electronics in Agriculture, vol. 211, p. 107956, 2023, doi: 10.1016/j.compag.2023.107956.

[3] V. Bidve, S. Mane, and P. Tamkhade, "Weed detection by using image processing," Indonesian Journal of Electrical Engineering and Computer Science,
vol. 30, no. 1, pp. 341–349, Apr. 2023, doi: [10.11591/ijeecs.v30.i1.pp341-349](https://doi.org/10.11591/ijeecs.v30.i1.pp341-349).

[4] G. S. Mmbando, "Harnessing artificial intelligence and remote sensing in climate-smart agriculture: The current strategies needed for enhancing global
food security," Cogent Food & Agriculture, vol. 11, no. 1, 2025, doi: [10.1080/23311932.2025.2454354](https://doi.org/10.1080/23311932.2025.2454354).

[5] Z. Wu, Y. Chen, B. Zhao, and X. Kang, "Review of weed detection methods based on computer vision," Sensors, vol. 21, no. 11, p. 3647, 2021, doi:
10.3390/s21113647.

[6] J. Huang, X. Xia, Z. Diao, X. Li, S. Zhao, J. Zhang, B. Zhang, and G. Li, "A lightweight model for weed detection based on the improved YOLOv8s
network in maize fields," Agronomy, vol. 14, no. 12, p. 3062, 2024, doi: 10.3390/agronomy14123062.

[7] M. Jones, R. Faroux, J.-N. Paoli, T. Maillot, C. Gée, and S. Villette, "Unsupervised classification algorithm for early weed detection in row-crops by
combining spatial and spectral information," Remote Sensing, vol. 10, no. 5, p. 761, 2018, doi: [10.3390/rs10050761](https://doi.org/10.3390/rs10050761).
[8] M. Kumar, P. Verma, and S. Gupta, "Texture and shape features for grass weed classification using hyperspectral remote sensing images," in Proc. IEEE Int.
Conf. on Precision Agriculture and Remote Sensing, 2020, pp. 345-350, doi: 10.1109/ICPA.2020.8900132.

[9] J. Huang, X. Xia, Z. Diao, X. Li, S. Zhao, J. Zhang, B. Zhang, and G. Li, "A lightweight model for weed detection based on the improved YOLOv8s network
in maize fields," Agronomy, vol. 14, p. 3062, 2024, doi: 10.3390/agronomy14123062.

[10] H. Pathak, C. Igathinathane, K. Howatt, and Z. Zhang, "Machine learning and handcrafted image processing methods for classifying common weeds in corn
field," Smart Agricultural Technology, vol. 5, p. 100249, 2023, doi: 10.1016/j.atech.2023.100249.

[11] F. D. Adhinata, Wahyono, and R. Sumiharto, "A comprehensive survey on weed and crop classification using machine learning and deep learning," Artificial
Intelligence in Agriculture, vol. 13, pp. 45–63, 2024, doi: 10.1016/j.aiia.2024.06.005.

[12] E. C. Rodríguez-Garlito, A. Paz-Gallardo and A. Plaza, "Automatic Detection of Aquatic Weeds: A Case Study in the Guadiana River, Spain," in IEEE
Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 15, pp. 8567-8585, 2022, doi: 10.1109/JSTARS.2022.3210373.

[13] A. Bakhshipour, "Cascading Feature Filtering and Boosting Algorithm for Plant Type Classification Based on Image Features," in IEEE Access, vol. 9, pp.
82021-82030, 2021, doi: 10.1109/ACCESS.2021.3086269.

[14] V. Singh, D. Singh and H. Kumar, "Efficient Application of Deep Neural Networks for Identifying Small and Multiple Weed Patches Using Drone Images,"
in IEEE Access, vol. 12, pp. 71982-71996, 2024, doi: 10.1109/ACCESS.2024.3402213.
[15] A. Naveed et al., "Saliency-Based Semantic Weeds Detection and Classification Using UAV Multispectral Imaging," in IEEE Access, vol. 11, pp. 11991-
12003, 2023, doi: 10.1109/ACCESS.2023.3242604.

[16] X. Jin, J. Che and Y. Chen, "Weed Identification Using Deep Learning and Image Processing in Vegetable Plantation," in IEEE Access, vol. 9, pp. 10940-
10950, 2021, doi: 10.1109/ACCESS.2021.3050296.

[17] Sowmya and S. Bhat "A comprehensive review of weed detection through advanced image processing and deep learning," Publication, vol. 4, pp. 59–66,
2024, doi: 10.5281/zenodo.10685044.

[18] M. S. Patel and R. Pillai R, "Design of an autonomous weed removal system," Proceedings of the 2019 4th International Conference on Advances in
Robotics, 2019, doi: 10.1145/3352593.3352674.

[19] H. Chegini, F. Beltran, and A. Mahanti, "Designing and developing a weed detection model for California Thistle," ACM Transactions on Internet
Technology, vol. 23, 2022, doi: 10.1145/3544491.

[20] S. T., S. T., S. G. G.S., S. S. and R. Kumaraswamy, "Performance Comparison of Weed Detection Algorithms," 2019 International Conference on
Communication and Signal Processing (ICCSP), Chennai, India, 2019, pp. 0843-0847, doi: 10.1109/ICCSP.2019.8698094.