Weed detection in soybean crop field using CNN

Abstract

This research contributes towards addressing the significant challenge of manual labor in weed detection and emphasizes the need for the development of an efficient system using digital image processing and deep learning techniques. The primary aim is to detect broadleaf weeds in soybean crops, leveraging the capabilities of Convolutional Neural Networks (CNNs) for image representation and pattern recognition. The study adopts an experimental research methodology, collecting a substantial dataset comprising 7,000 soybean images and 17,000 images for broadleaf weed, soil, and grass classes. To ensure balanced training, 2,000 images from each class are used. Digital image preprocessing is used to get the data gathered ready for analysis. The acquired image data passes through cleaning and data filtering. Additionally, in order to give class labels and identify regions of interest for crop and weed identification, relevant labelling and annotation processes are carried out. The classification using convolutional neural network (CNN) involves four classes: broadleaf weed, soil, soybean, and grass. Utilizing 80% of the total dataset for training and the remaining 20% for testing, the experimental results demonstrate the efficiency and accuracy of the proposed model. Specifically, the model achieves a weed detection accuracy of 94.46%, indicating its promising potential for real-time weed detection in soybean fields. This accomplishment is a crucial step towards mitigating the reliance on manual labor, enabling timely and accurate weed identification to enhance crop management. Looking ahead, future work is recommended to expand the model's recognition capabilities to include other weed species and roots. The incorporation of a larger dataset with diverse images would further enhance the model's robustness and generalization to different environmental conditions. The ultimate goal is to develop a comprehensive and versatile model that can contribute significantly to precision agriculture by not only identifying broadleaf weeds but also expanding its scope to encompass various weed types and agricultural challenges.