GCS-YOLO: A lightweight strawberry disease detection algorithm based on improved YOLOv8

In order to solve the problems of low efficiency and accuracy of manual strawberry disease identification, and the large number of parameters and high computational requirements of CNN-based target detection models, which make it difficult to be deployed on a large scale in embedded devices with limited computational performance and storage resources, a lightweight strawberry disease detection algorithm, GCS-YOLO, based on YOLOv8s, is proposed. First, the standard convolution of the backbone part is replaced with a more lightweight ghost convolution (GhostConv) to reduce the size of the model parameters, and a GC2f module is designed for the backbone and neck parts. convolution is replaced by the more lightweight GhostConv to reduce the size of the model parameters, and the GC2f module is designed for the backbone and neck parts. Coordinate Attention (CA) was added to the backbone to enhance the backbone's ability to extract strawberry disease features. Finally, the SIoU loss function was used for the bounding box regression loss of the model. The experimental results on the dataset we used show that the average accuracy (mAP50) of the improved GCS-YOLO is 1.6% higher than the original model under the same training conditions, the Precision is 4% higher, and the number of parameters and model size are reduced to 6.6M and 13.0M, respectively. the improved model has a higher detection accuracy than the original model, and has a lower hardware requirement, which provides a new way to detect strawberry diseases. lower, providing a new method for strawberry disease identification.