Abstract: An image enhancement and improvement of DeepLabv3+ drawing recognition and correction method is proposed to replace the teacher to correct drawing assignments. Image pre- processing was accomplished by image rotation correction, question number recognition, and image contrast enhancement. The original DeepLabv3+ was improved by adopting EfficientNetv2 to enhance the model training efficiency, adding the convolutional attention mechanism to the shallow semantic information, and enhancing the model's ability to recognize image line shapes with the hybrid null convolution of feature stacking. The skeleton extraction algorithm was used to retain the main part of the line, and the seed filling method was used to obtain the coordinates, and the sketch was reconstructed based on the coordinate values. The reconstructed sketch was paired with the standard answer to realize automatic correction of homework. The experiment proves that the method is accurate, efficient and has practical application value.