Objective: To solve the problems of low detection accuracy and large number of model parameters in existing tomato maturity detection methods. Methods: Based on the tomato image acquisition system, an improved YOLOv4 model was proposed for automatic detection of tomato maturity. Introducing the lightweight network MobileNetv3 network into the model to replace the CSPParkNet53 network, reducing model complexity. Using average pooling instead of maximum pooling in the SPP module improved the algorithm's detection accuracy for small targets. Introduced attention mechanism CBAM in the upsampling process to enhance the fusion ability of deep and shallow features, and verified the feasibility of the proposed model through experiments. Results: Compared with conventional methods, the experimental method had higher detection mAP values and operational efficiency in tomato maturity detection, and the model parameter quantity was relatively small, the mAP value was 92.50%, the detection speed was 37.1 FPS, and the model parameter quantity was 48 M. Conclusion: This tomato maturity detection method can effectively reduce model parameters and detection time, and has a high detection mAP value.