A Lightweight YOLOv8-DGS Model with ESP32-CAM-based Image Acquisition for Strawberry Ripeness Detection
Tian Shuo, Wang Guizhen, Fu Tengfei, Yang Lishan, Sun QunObjective:
Addressing the issues of dependence on manual labour, low detection efficiency, and high hardware costs in existing intelligent recognition systems for automated strawberry picking, yield estimation, and growth monitoring, this paper designs a low-cost, low-power, and high-precision intelligent detection scheme for strawberry ripeness, providing deployable technical support for practical agricultural production.
Methods:
A strawberry maturity detection system based on ESP32-CAM is developed with an “image acquisition–wireless transmission–model inference–visualization” architecture. The front end captures images via ESP32-CAM, while a PyTorch server deploys an improved YOLOv8n model (YOLOv8-DGS). DWConv reduces computational complexity, GhostConv suppresses redundant features, and the parameter-free SimAM attention mechanism enhances key feature representation.
Results:
The experimental results conducted on a self-built strawberry maturity dataset show that the improved YOLOv8-DGS model achieves a mAP@50 of 0.800, representing an improvement of approximately 0.4% compared to the original YOLOv8n model. At the same time, the number of model parameters is reduced by approximately 11%. The experiment verifies that the proposed model achieves lighter performance while maintaining high detection accuracy.
Discussion:
The experimental results demonstrate that by integrating dual lightweight convolutional modules and a parameter-free attention mechanism within a unified network structure, detection accuracy can be effectively enhanced with minimal increase in model complexity. This approach ensures real-time performance while reducing dependence on hardware computing power, thereby enhancing the practicality and scalability of the system in agricultural greenhouse scenarios.
Conclusion:
This paper proposes a strawberry maturity detection system based on ESP32-CAM and an improved YOLOv8-DGS model, achieving low-cost, low-power real-time image acquisition and wireless transmission. Rapid, accurate maturity recognition is performed on the server side. The research results show that this solution has strong application prospects in intelligent strawberry maturity detection and can serve as a reference for the automatic monitoring and intelligent management of fruits and vegetables in facility agriculture.