Abstract Tool wear state recognition is a challenging problem because the various types of signals corresponding to different wear states have similar features, making classification difficult. Convolutional neural networks are widely used in the field of tool wear state recognition. To address this problem, this study proposes a milling cutter wear state monitoring model that combines KANS and deep residual networks (ResNet). The traditional ResNet-34’s top linear classifier was replaced with a nonlinear convolutional classifier including _top_kan, and the data were preprocessed using continuous wavelet transform (CWT) to enhance the model’s immunity to interference and feature characterization. The experimental results based on the PHM dataset show that the improved KANS-ResNet-34 model improves accuracy by 1.07% compared to ResNet-34, making it comparable to ResNet-50, while its computation time is only 1/33.68 of the latter. This significantly improves computational efficiency, reduces the pressure on hardware resources, and provides an effective tool wear state recognition solution. Keywords: deep learning; tool wear monitoring; KAN networks; residual networks; continuous wavelet transform