A method for diagnosing faults in milling machines (MM) using defect-sensitive scalograms (DSS) and convolutional neural networks (CNN) is proposed in this work. Deep learning (DL) techniques, specifically the extraction of health-sensitive features from the scalograms of continuous wavelet transform (CWT), have gained popularity for milling machine defect diagnosis. However, acoustic emission signals (AE) from milling machines often contain both defect-related information and unnecessary background noise, which can hinder the ability of deep learning models to autonomously extract relevant health-sensitive features. To address this issue, a novel approach involving defect-sensitive scalograms is proposed. These scalograms improve the visualization of defect-related color intensity changes and minimize noise through the use of Gaussian and Sobel image filters. The method identifies the condition of the milling machine by extracting defect-relevant data from these defect-sensitive scalograms and classifying them using convolutional neural networks. When evaluated using real-world milling machine acoustic emission data, the proposed method achieved a classification accuracy of 97.1%, surpassing the performance of existing techniques.