An Improved YOLOv5 Object Detection Model in BALATRO Built on Tensorflow2

To improve the recognition efficiency of game interfaces, this paper proposes an improved game interface detection method based on the improved YOLO v5 network. The proposed method uses CSPDarknet backbone feature extraction network and FPN feature pyramid to enhance the network's perception of small objects, and multi-attribute decoupled detection header to improve the efficiency of playing card classification. A dataset of multi-type cards collected from the game interface was built. The detection model is trained based on these data. Experiments show that the lightweight deployment of the model can provide fast and accurate prediction results with good practicality and portability.