<p>This study enhances the machining accuracy of arc-shaped CVD diamond tools by developing a novel method that utilizes acoustic emission technology. This approach combines spatial energy variation characteristics with localized frequency-domain features obtained through wavelet transform. Initially, the acoustic emission signals are processed using a segmentation method, and the root mean square (RMS) sequence of these segmented signals represents the spatial energy distribution. Subsequently, a wavelet transform is applied to the acoustic emission signals to extract frequency components relevant to grinding state recognition. To improve the recognition capability, this study introduces a dual-channel convolutional neural network model. In this model, both the RMS sequence and the wavelet-extracted components are concurrently fed into a dual-branch convolutional neural network. The model’s efficacy in capturing critical features is augmented by incorporating channel attention modules and a multi-feature fusion module. Experimental outcomes confirm that the proposed method attains a state recognition accuracy of 94.19% during the grinding process. This method surpasses other methods in performance, demonstrating its ability to effectively fuse multi-scale features from acoustic emission signals, thereby improving the accuracy of grinding state identification.</p>