Tomato Maturity Recognition Model Based on Improved YOLOv5 in Greenhouse

Due to the dense distribution of tomato fruit with similar morphologies and colors, it is difficult recognize maturity stages when harvested. In this study, a recognition model, YOLOv5s-tomato, proposed based on improved YOLOv5 four types different stages: mature green, breaker, pink, red. Tomato datasets were established using images collected at maturing in greenhouse. The small-target detection performance model was by Mosaic data enhancement. Focus Cross Stage Partial Network (CSPNet) adopted improve speed network training reasoning. Efficient IoU (EIoU) loss used replace Complete (CIoU) optimize regression process prediction box. Finally, algorithm compared original dataset. experiment results show that YOLOv5s-tomato reaches precision 95.58% mean Average Precision (mAP) 97.42%; they are 0.11% 0.66%, respectively, YOLOv5s model. per-image 9.2 ms, size 23.9 MB. can effectively solve problem low accuracy for occluded tomatoes, also meet requirements greenhouses, making suitable deployment mobile agricultural devices provide technical support precise operation tomato-picking machines.