Comparative study of YOLOv8 and YOLO-NAS for agriculture application

Agricultural tasks have significantly improved as a result of ongoing machine learning (ML) improvements. Deep learning (DL), which has a significant capacity for extracting high-dimensional features from fruit images, is widely applied to the automated detection and harvesting of fruits. In the fields of fruit recognition and automated harvesting, Convolutional Neural Networks (CNNs) have demonstrated the ability to attain speed and accuracy levels that rival human performance. This article compares the performance of YOLOv8m with YOLO-NASl for grapes detection. In this research, the YOLOv8 and YOLO-nas object detection models, including their different scales, were trained using a publicly available Embrapa WGISD dataset. The dataset consists of 300 digital images of grapes growing in vineyard settings, and it includes a total of 4,432 annotations. The performance of the YOLOv8m and YOLO-NASl model were evaluated using metrics such as recall, precision, and the mean average precision (mAP@50). In the subset of test data, YOLOv8m achieved the top overall performance, with a precision (0.855), mAP@50 (0.885), and recall (0.827), while best recall was obtained from YOLO-NASl (0.934).