A construction waste classification method based on a CNN comprises the following steps that 1, construction waste pictures are collected, a construction waste material data set is self-made, and data set samples are enriched through data enhancement; step 2) using the lightweight self-made convolutional neural network model for construction waste classification: inputting the preprocessed pictures into the trained network model, and finally outputting a group of probability values, the category with the maximum probability value being a result category; 3) in the CNN, performing convolution operation on the image through a convolution kernel, extracting image feature information, in construction waste classification application, noise in data set image information is inevitable, and in order to reduce noise transmission, using a maximum pooling mode, and reducing feature image dimensions; and (4) an AdamOptimizer optimizer is used for optimizing model parameters, and therefore accurate classification of four types of construction waste materials including red bricks, foam, concrete and PVC is achieved. According to the invention, accurate classification of four types of construction waste materials is realized.