In order to obtain accurate information on the degree of plaque development in patients' blood vessels, and to assist clinicians in judging and recognizing atherosclerotic areas, a deep learning-based study of intravascular ultrasound atherosclerotic plaque development was performed (CPCA). First, different types of ROIs are extracted for plaque images. Secondly, according to different ROI regions, the size of the sliding neighborhood block is determined, and the central pixel traverses the plaque region to obtain a small image slice of the plaque developing region. Then, based on PCAnet based on principal component analysis vector as convolution kernel, a clustering PCA network is designed to cluster small image slices and calculate principal component vectors by category to generate multiple sets of convolution kernels. The multi-plaque visualization feature enables the input image to adaptively select the feature extractor to achieve classification recognition of the degree of plaque development. The result of manual labeling by doctors is taken as the standard true value. The experimental results show that the proposed algorithm can effectively extract the features of plaque developed images and achieve high-efficiency recognition of plaque development. INDEX TERMS Deep learning, plaque, degree of development, recognition.