YOLO-VDS: accurate detection of strawberry developmental stages for embedded agricultural robots

Abstract Detecting the various developmental stages of strawberries in their natural environment is crucial for modern agricultural robots. Existing methods focus on fruit detection but overlook stage classification. Moreover, they often require substantial computational resources, making them unsuitable for small, low-power embedded platforms. To address this issue, we propose YOLO-VDS, a lightweight model based on YOLOv5s and optimized for embedded platforms. We introduce the Inverse Residual Bottleneck with 3 Convolutions (IRBC3) module to enhance feature extraction capabilities and reduce the model computation. Additionally, we improve the feature extraction and representation capabilities by incorporating the Efficient Channel Attention (ECA) module into the backbone. Experiments on the Strawberry-DS dataset show that YOLO-VDS significantly outperforms other similar algorithms such as YOLOv5s and YOLOv4-v11. Compared to YOLOv5s, accuracy improves by 5.8%, [email protected] increases by 7.7%, and model parameters are reduced by 24.29%. When deployed on a Jetson TX2 NX, YOLO-VDS reaches 19.2 FPS after TensorRT acceleration, demonstrating its suitability for vision-guided harvesting robots and edge computing applications.