Weed Detection in Cotton Fields Using YOLOv5: Prototype Development and Deployment

Weed control is a critical task for any crop production, as weeds are a substantial threat to crop yields. To ensure effective weed control with site- and species-specific herbicides, it is essential to classify and localize them in the field precisely. In this study, we employ deep learning and machine vision technology to address this issue. We train a state-of-the-art deep object detector, the YOLOvS model, on images collected from real cotton fields to detect weeds. We also apply various context-based data augmentation techniques to enhance the model's robustness. The trained YOLOv5 model has been deployed on a cost-efficient single-board computer to develop a prototype that can be used in the field. Additionally, hardware has been developed to generate signals that will trigger a herbicide sprayer in the event of weed detection. The prototype can detect cotton weeds in real-time with a high confidence score and generate the signals precisely. This study will serve as a valuable resource for developing next-generation prototypes and eventually industrial-scale field-deployable equipment for weed detection.