BCP-YOLO: A Lightweight Industrial Tomato Detection Method for UAV Inspection

Timely, accurate, and comprehensive detection of industrial tomato ripening is critical for farmers to capture the best harvest timing as well as optimize yield and quality. Aiming at the problem that deep learning models are complex and difficult to be deployed on performance- and resource-constrained unmanned aerial vehicle (UAV) devices, this paper proposes a lightweight industrial tomato detection method based on YOLOv8n. First, a lightweight bi-directional feature pyramid network with learnable weights is introduced in the model neck to learn the importance of different input features and provide strong semantic information for the model; second, a C2f_EMBC module based on the attention mechanism is proposed to enable the model to apply more attention to important features; finally, a channel pruning algorithm is used to further reduce the size and complexity of the improved model and enhance the model's detection speed. The test results show that the mAP value of the model is 95.3%, which is 0.3% higher than the original model, the number of parameters and computation of the model are reduced by 66.06% and 58.02%, respectively, and the detection speed is improved by 17.6%, and the model size is only 2.3 MB. comparing with 10 methods, our method shows the optimal performance. The research method provides a concrete case for the application of deep learning in agriculture, which can provide technical support for intelligent and automated production in agriculture.