Based on improved YOLOv7 tomato fruit recognition method in natural environment

Traditional methods for tomato fruit recognition and detection face challenges in natural environments, including low recognition accuracy and slow processing speeds. To achieve automated and accurate recognition of tomato fruits in complex environments, thereby facilitating automatic fruit picking, this study proposes a method based on an enhanced YOLOv7 model for tomato fruit recognition. This approach modifies the original YOLOv7 model by integrating SEnet structures into the ELAN module within the backbone network, allowing for the assessment of the importance of different channel features. Additionally, a global attention GAM module is incorporated at the Neck end of the model to improve the network's feature representation capabilities. Experimental validation demonstrates that the improved YOLOv7 model performs satisfactorily on the test set, achieving a mean accuracy of 98.9%. Compared to the original YOLOv7 model, the enhanced version exhibits a mean average precision increase of 6.2 percentage points. The comprehensive results indicate that the improved YOLOv7 model offers robust technical support and insights for the realization of automated tomato fruit picking.