A deep learning-based recognition and classification model named Tenon Grinding Burn Net （TenonGBNet） is proposed to address the issues of misdiagnosis and missed detection in visual inspection of grinding burns on nickel-based Superalloy blades. The K4125 nickel-based superalloy blades are chosen as the target， and through grinding burn tests and specimen organization inspection， a set of classification standards and corresponding image collection for different degrees of blade tenon grinding burns are obtained. Then， ODConv dynamic convolution is employed to fuse Inception V2 modules and the Coordinate Attention mechanism to enhance the model's feature extraction capability while ensuring the model is lightweight. Finally， a fully connected layer is employed for identification and classification. Experimental results indicate that， compared with four other classical classification models， TenonGBNet achieves an average classification accuracy of 96.50% while maintaining a minor model complexity and parameter count. Additionally， the classification accuracy for each burn level exceeds 95%.