Apple detection in complex environment based on improved YOLOv8n

Abstract Due to the complexity of apple growth environment and the large number of fruits, the
existing target detection algorithms are difficult to achieve balance in fruit detection accuracy and model lightweight.An apple detection algorithm based on Faster-EMA Block and Lightweight neck in complex environment is proposed.Firstly, the Bottleneck of Faster-EMA
Block replacing C2f of Backbone is constructed by using PConv (Partial Convolution) and EMA (Efficient Multi-Scale Attention) attention mechanism. Secondly, an efficient feature fusion technology is adopted, especially the lightweight processing of neck features, so that the model can maintain a high level of accuracy while reducing the burden of parameters.Then, in order to deepen the model's ability in vision detection, we integrated the detection layer with unobvious apple features to capture the apple shape more accurately and improve the detection accuracy. Finally, to evaluate the matching degree between the prediction results and the real shape more accurately, the Shape IoU is introduced.
Experiments show that, compared with the YOLOv8 baseline network, mAP50 was increased by 1.5%, mAP50-95 was increased by 1.7%, and params was reduced by 20% on the hybrid apple dataset. The improved network has a better detection effect on apples,which lays the foundation for the intelligent picking of apples.