Edge-Cloud Collaborative Multi-Target Detection and Growth Prediction for Greenhouse Strawberries Based on YOLOv8 and Multimodal Fusion

Abstract

To address the problems of low manual inspection efficiency and delayed disease detection in greenhouse strawberry cultivation, this paper constructed a multi-target visual detection model based on the YOLOv8 deep learning algorithm, which simultaneously achieved accurate identification of four types of targets: strawberry ripeness, diseases, pests and growth stages. The research dataset was built by combining images continuously collected from field greenhouses and public agricultural datasets, with a total of 17,000 images. Controlled experiments were conducted after image preprocessing, manual annotation and model training. The lightweight model was deployed on the K230 embedded chip to realize local edge inference, and integrated with environmental sensing data through the Vision-Environment-Terminal Network (VET-Net) multimodal fusion architecture to achieve 7-day crop growth trend prediction. Experimental results showed that the identification accuracy of all four visual targets exceeded 94%, and the end-to-cloud data transmission delay was controlled within 2 seconds. The whole scheme was deployed on an autonomous inspection robot, and farmers can obtain real-time detection results through a WeChat Mini Program. The system effectively reduced manual inspection workload and enabled early disease risk prevention, which possesses good practical application value for reducing strawberry planting costs and increasing yield per unit area.