YOLO-Wheat: A Wheat Disease Detection Algorithm Improved by YOLOv8s

The objective of this research is to create an algorithm for identifying wheat diseases in their natural habitat. With its adaptability to small targets and intricate surroundings, the algorithm is expected to offer a precise foundation for scientific management and disease prevention. First, a total of 3622 original datasets of wheat disease photos were acquired by taking wheat photographs in an agricultural environment over various periods. Second, YOLO-Wheat, a detection method, is suggested for the properties of infected wheat ears and leaves that share a limited detection target area and comparable look, texture, color, and other aspects. In order to improve the extraction of illness characteristics and get a wider sensory field for the input features, the algorithm makes use of the new C2f-DCN module and the SCNet attention mechanism. Moreover, it enhances the model's capacity to extract information from distorted objects by learning the offset and weighting. Moreover, to improve the identification of minor illnesses, the design has enlarged the detection layer for tiny targets, modified the detecting head, and optimized the loss function. All of these changes have improved the accuracy of minor disease detection. On the experimental dataset, the YOLO-Wheat method earned a mAP@0.5 of 93.28%, which is 12% better than the original model. The suggested approach shows a 47% performance gain over the previous model, while maintaining a 23.94 MB smaller algorithm size. The results of this study show that the approach may greatly increase the model's capacity to extract features from tiny target pictures as well as the robustness of crop disease detection. As a result, the technique may be precisely and successfully used in real-world crop disease detection settings.