YOLO-ASFF: A Model for Detecting Tomatoes of Different Maturities Based on Improved YOLO v5s

In agricultural production, accurate detection of tomatoes of different maturities is crucial to control the picking time and improve grading efficiency. However, the orchard environment is complex, which brings many challenges to detecting tomatoes of different maturities, such as variable light conditions, lush branches and leaves, fruit occlusion, tomatoes with different sizes, etc. Traditional methods rely on manual experience, which is not only time-consuming but also has limited accuracy. To this end, the YOLO-ASFF model is proposed to improve the detection accuracy of tomatoes of different maturities in complex orchard environments in this work. On the one hand, the model uses the extended intersection over union (EIoU) to optimize the bounding box positioning to ensure that tomatoes of different sizes and maturities could be accurately framed. On the other hand, the adaptive spatial feature fusion (ASFF) module was integrated to enhance the fusion ability of multi-scale features, so that the model could capture feature information more comprehensively when detecting tomatoes with different maturities. The experimental results show that the YOLO-ASFF model performs better than other models, including Faster R-CNN, EfficientNet, YOLO v5s, and YOLO v11n, in detecting tomatoes of different maturities, with an accuracy of 0.885, a recall rate of 0.835, a mean average precision of 0.850, and an Fl-Score of 0.850. In short, YOLO-ASFF can provide an effective method for accurately identifying tomatoes with different maturities, which can help reduce labor costs and improve picking efficiency.