A Lightweight Bamboo Shoot Segmentation Method Based on Deep Convolutional Network

The precise segmentation of bamboo shoots can realize intelligent monitoring of bamboo shoot growth, which is of significant significance to the optimization of bamboo garden production management. This study adopts a lightweight method based on deep convolutional network for bamboo shoot segmentation. First, an experiment was designed to collect images of bamboo shoots in the wild, and a training set and a test set were constructed through manual labeling. Then, the YOLOv8 deep convolutional network model was used as the basic model for bamboo shoot segmentation, and model training and verification were conducted. Finally, this study designed two sets of comparative experiments to effectively verify the accuracy and efficiency of the method. Experimental results show that the detection and segmentation mAP values of the proposed method on the bamboo shoot test set are 0.927 and 0.931 respectively, which are 0.1% and 0.4% higher than the YOLOv8_x network respectively. In addition, the FPS, params and Model_Size of this method are 169, 43.8MB and 92.3MB respectively, which are 51, 24.6MB and 51.6MB higher than the YOLOv8_x network respectively. Experimental results show that this method has high accuracy and real-time performance in bamboo shoot segmentation, and provides important technical support for bamboo orchard production management.