Abstract
As global food demand increases and environmental concerns grow, improving weed management has become essential for sustainable agriculture. Weeds significantly reduce crop yields, and traditional control methods often depend on widespread herbicide application, increasing costs and ecological impact. This study explores the use of deep learning to improve crop-versus-weed detection and multi-species weed classification, supporting more precise and efficient agricultural practices. The research utilizes two publicly available datasets from Kaggle: one for binary crop-versus-weed classification and another covering 12 weed species. Four preprocessing techniques (Baseline, Region of Interest (ROI), CLAHE, and ROI + CLAHE) were implemented to enhance plant features and reduce background noise. To address class imbalance, datasets were oversampled prior to training. Multiple convolutional neural network architectures were evaluated, including ResNet, VGG16, DenseNet201, MobileNet, and YOLO. Model performance was assessed using accuracy, precision, recall, and F1 score. Results indicate that both preprocessing strategy and model architecture significantly influence performance. In binary classification, top-performing models achieved 96–98% accuracy, with ROI + CLAHE providing the greatest improvement. Although the 12-class task was more complex, balanced data and enhanced preprocessing led to strong and consistent multiclass results. Overall, the findings demonstrate that effective preprocessing combined with robust deep learning models can significantly improve weed detection. For Kansas, where agriculture is a major economic driver, this approach can support precision farming, reduce herbicide usage, lower production costs, and promote environmentally sustainable crop management.
Faculty Advisor
Anas Hourani
Department/Program
Computer Science
Submission Type
online only poster
Date
3-4-2026
Rights
Copyright the Author(s)
Recommended Citation
Velo Castaneda, Ivan A.
(2026)
"Deep Learning-Based Crop and Weed Detection System for Precision Agriculture: Binary and Multiclass Classification Approaches,"
SACAD: Scholarly Activities: Vol. 2026, Article 18.
Available at:
https://scholars.fhsu.edu/sacad/vol2026/iss2026/18
