Detection method of cherry tomato fruit ripeness based on SSPENet

DUAN Xin'e; ZHANG Zhiwang; ZHOU Qingxing

doi:10.16861/j.cnki.zggc.2025.0214

Abstract：To achieve rapid and accurate identification of cherry tomato fruit ripeness in greenhouse environment,this study proposed a ripeness detection algorithm based on the stereoscopic spatial pyramid attention network(SSPENet).First,a spatial stereoscopic attention mechanism(SSAM)was constructed to enhance the perception of fruit features by adaptively focusing on key regions.Second,a local attention pyramid module(LAPM)was incorporated into the neck network to strengthen the feature fusion of small-scale cherry tomato,thereby improving detection accuracy for small-scale targets.Finally,an efficient geometric regression loss function(LEGR)was proposed to optimize the geometric properties of bounding boxes,further improving the localization accuracy for small-scale cherry tomatoes.The experimental results showed that SSPENet achieved 96.1％mAP on the cherry tomato ripeness dataset,representing a 5.1 percentage point improvement over the baseline model,with an inference speed of 94.7 frames per second.This achieved a good balance between detection accuracy and computational efficiency.This study provides an efficient and scalable technical solution for cherry tomato ripeness detection in greenhouse environment,with broad application prospects.