Lightweight Detection Algorithm of Kiwifruit Based on Improved YOLOX-S

Considering the high requirements of current kiwifruit picking recognition systems for mobile devices, including small number available features image targets and small-scale aggregation, an enhanced YOLOX-S target detection algorithm robots is proposed in this study. This involved designing a new multi-scale feature integration structure which, with aim providing lightweight model, maps used detecting large YOLOX model are eliminated, map sampled through nearest neighbor values, superficial spliced final features, gradient SiLU activation function perturbed, loss at output optimized. The experimental results show that, compared original YOLOX-S, improved average precision (AP) images by 6.52%, reduced parameters 44.8%, speed 63.9%. Hence, its outstanding effectiveness relatively light weight, capable effectively data support 3D positioning automated kiwifruit. It may also successfully provide solutions similar fields related to detection.