Answers are grounded in this paper and cite their sources.
Strawberry-picking robots have played an important role in improving strawberry production efficiency. However, the restricted computation ability of these edge devices and the high complexity of the strawberry growth environment and conditions are not conducive to the practical application of the strawberry detection network. Against this backdrop, this study proposes an improved YOLOv11 network called MVYOLO-Tiny, which is deployed on Android-based smartphones and Jetson Xavier NX for real-time detection. To build a robust model, a dataset comprising strawberry images across different climes, scenarios, and growth stages is constructed. Three main improvements are introduced into the proposed model MVYOLO-Tiny: replacing the C3k2 models in the backbone with the C2fA model with MobileViTv2 blocks, removing the modules of the P5 layer in the neck and head, and incorporating the lightweight modules, VoVGSCSP, along with the cross-layer residual connection (CRC) and the weighted concatenating mechanism (WCM) in the BiFPN architecture into the neck. Compared to YOLOv11, MVYOLO-Tiny achieves 0.2%, 0.8%, and 0.6% improvement in P, R, and mAP@0.5 respectively, reaching 90.1%, 88.5%, and 94.0%. Simultaneously, the parameter count and GFLOPs decrease by 54.1%, and 14.1% respectively. The analysis additionally demonstrates its enhanced detection capabilities for occluded, overlapping, and small-sized targets in complex scenarios. The Android-based APP exhibits its lower computational costs. On Jetson Xavier NX, MVYOLO-Tiny achieves a speed increase of more than 20%. These make it hold the potential for the deployment of strawberry-picking robots, thereby contributing to advancing the application and facilitating the promotion of smart agriculture in strawberry cultivation.
No takes yet. Share an insight, caveat, or question.
Li et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: