Multi-target small object identification method based on category loss and difference detection

The invention discloses a multi-target small object recognition method based on category loss and difference detection, which comprises the following steps of: acquiring a chessboard initial image, calibrating a chessboard image coordinate, and segmenting the chessboard image according to the coordinate; preprocessing the segmented chessboard image, classifying the processed image by using the built deep learning network model, and converting a classification result into chess information; a chessboard image P1 is collected, and a chessboard image P2 is collected again after a chess player performs walking chess; according to the chessboard image P1 and the chessboard image P2, a difference value detection method is adopted for conducting reasonability judgment on the starting point and the ending point of chess moving, and chess information is updated. The method realizes detection and identification of Chinese chess from opening to each walking chess, has the advantages of high speed, high robustness and th