Abstract

For the complete implementation of site-specific weed management, which is currently a major challenge in modern agriculture, precise weed mapping is essential for sustainability, efficiency, and the maintenance of high crop yields and less chemically polluted agricultural lands. In this study, the robustness of the training epochs of the Convolutional Neural Network (CNN) model You Only Look Once (YOLO) v5s was evaluated for the creation of an automatic crop and weed classification using UAV images. The pictures were explained utilizing a jumping box and they were prepared on Google collaboratory north of 100, 300, 500, 600, 700, and 1000 ages. Sugarcane (Saccharum officinarum), banana trees (Musa), spinach (Spinacia oleracea), pepper (Capsicum), and weeds were all identified and categorized by the model. The model was trained over a number of epochs to find the best performance on the test set. When the test performance (classification accuracy, precision, and recall) started to drop, training was stopped. The result shows that the classifier's performance improved significantly as the number of training epochs increased, typically from 100 to 600. When the number of epochs was increased to 700, classification accuracy, weed precision, and recall were recorded at 65, 43, and 43%, respectively, compared to 67, 78, and 34% at 600 epochs, respectively. In the meantime, a slight decline was observed. When the epoch was increased to 1000, classification accuracy, weed precision, and recall of 65 percent, 45 percent, and 40 percent, respectively, were achieved, but this decline persisted. The findings revealed that the YOLOv5s training epoch has a significant impact on the model's robustness in automatic crop and weep classification, with 600 being the optimal epoch.