Land Cover Classification for Identifying the Agriculture Fields Using Versions of YOLO V8

Accurate identification and classification of agricultural fields is essential for analyzing crop growth, agricultural resource management, and supporting decision-making in precision farming. The current study uses the remote sensing images to classify the agricultural lands using the You Only Look Once (YOLO) V8 approach. The dataset consists of multiple classes such as forests, river, sealake, highway, pasture, residential, industrial, and permanent crop. Various versions of YOLO V8, namely the nano, small, and medium versions of the deep learning model are used in the classification of the sensor images. The impact of the hyperparameters such as the number of epochs, optimizers, learning rate, momentum, and weight decay are analyzed across all the versions of YOLO V8. The experimental outcome demonstrates the performance of each model concerning the accuracy, precision, and recall. Based on the experimental performance, the YOLO V8 model is efficient in precisely recognizing various land cover, which result in offering a scalable and efficient approach for real-time agricultural field identification. From the experimental results, it can be stated that the medium variant has achieved the highest top_1 accuracy value of 99% at 50 epochs while nano and small variant achieves accuracy values of 98.60% and 98.50% at the same epoch values, respectively.