Research on Tomato Flower and Fruit Recognition and Detection Methods Based on Image Vision

This study proposes an image vision-based recognition and detection method utilizing deep learning technologies for tomato flowers and fruits. A high-quality image dataset comprising approximately 10,000 annotated images of tomato flowers and fruits at different growth stages was constructed, incorporating diverse environmental and lighting conditions. Data augmentation techniques such as random rotation, brightness adjustment, and scaling were applied to enhance dataset robustness. A novel detection model was developed, integrating a Spatial Pyramid Pooling Fast (SPPF) module and a Dynamic Downsampling (Dysample) module to extract multi-scale features and optimize small object detection. The model was trained and validated using the PyTorch framework on a high-performance computing environment, with hyperparameters fine-tuned for accuracy and efficiency. Experimental results demonstrate the model's superior performance in terms of detection accuracy, recall rate, and inference speed compared to state-of-the-art methods such as YOLOv5 and Faster R-CNN. Specifically, the model achieved a mean average precision (mAP) of 92.3% at IoU 0.5 and a recall rate of 87.6%, while maintaining a lightweight design suitable for real-time agricultural applications. This study provides a methodological foundation for applying similar detection strategies to other crops.