Detection of Cotton Plants Using the YOLOv7 Deep Learning Model

Abstract

Detecting objects in a real environment presents challenges due to fluctuating lighting and diverse object attributes. This study elucidates a deep learning (DL) object detection model for recognizing cotton plants. The DL-driven YOLOv7 model was developed for expedited recognition of cotton plants, leveraging its rapid inference capabilities. Employing a convolutional neural network (CNN), the object detection model was instantiated. The developed object detection was executed in Anaconda software to identify the cotton plants from input images. This advancement holds the potential to automate cotton farming tasks like precise applications, optimized pesticide usage, pest management, robotic harvesting, and targeted weeding. After 100 epochs, the YOLOv7 model achieved bounding box and object loss of 2.12% and 1.32%, respectively. Precision, recall, F1 score, and mAP were determined as 1.00 at 0.696 confidence, 097 at 0.000 confidence, 0.87 at 0.151 confidence, and 0.879 at 0.5 confidence, respectively.