Determination of the damage caused by some vineyard pests with deep learning techniques

Global warming caused by climate change causes various difficulties in agricultural production. One of the challenges in agricultural production is the increase in various pest populations. The increase in population seriously threatens agricultural products and significantly negatively affects productivity and quality. Most of the time, farmers have difficulty in identifying pests and their effects, which can lead to incorrect and excessive insecticide applications. Excessive use of insecticides threatens human health and causes environmental pollution, while increasing production costs and creating economic pressure on farmers. Therefore, early detection of pests and the damage they cause to plants is of critical importance for the sustainability of agricultural systems. In this study, a YOLOv8n-based model was improved to detect the damage of some important vineyard pests such as grapevine moth, thrips, two-spotted spider mite and vineyard leaf scab in vineyards using deep learning methods. The improved model is named AgDet-YOLO. In this model, the SPPF block is replaced by the GSPPF block, the SiLU activation function in the convolution layer is replaced by ReLU, and the convolution layers in the neck network of the model are replaced by the ghost convolution layer. Additionally, the Global Attention Mechanism (GAM) module was added to the backbone network before the GSPPF block. By changing the layer connections in the head network, the total number of parameters of the model was reduced from 3.2M to 2.9M. During the training process, while the overall class success of the improved model was 0.955 in the mAP@0.5 metric in the standard model, this value increased to 0.963 in the improved model. Additionally, performance comparisons were made with the improved model, YOLOv5n, YOLOv8n models and a different single-stage model, MobileNetv2-SSD.