Recognition and Detection of Persimmon in a Natural Environment Based on an Improved YOLOv5 Model

Accurate and rapid recognition of fruit is the guarantee intelligent persimmon picking. Given changes in light occlusion conditions a natural environment, this study developed detection method based on improved YOLOv5 model. This approach has several critical steps, including optimizing loss function traditional YOLOv5, combining centralized feature pyramid (CFP), integrating convolutional block attention module (CBAM), adding small target layer. Images ripe unripe persimmons were collected from trees. These images preprocessed to enhance contrast, they then extended by means image enhancement increase robustness network. To test proposed method, experiments, comparative conducted. From environment could be detected successfully using model, with accuracy rate reaching 92.69%, recall 94.05%, average 95.53%. Furthermore, comparison model performed better than single-shot multibox detector (SSD) models, improving while reducing leak false rate. findings provide some references for automatic picking persimmons.