Label noise identification method, system and equipment for vehicle style image data

The invention relates to a label noise identification method, system and device for vehicle style image data, and the method comprises the steps: employing a multi-task framework of a convolutional neural network, setting a sub-network structure corresponding to vehicle style label levels, carrying out the feature extraction of vehicle images at different levels, and carrying out the recognition of the label noise of the vehicle style image data based on the features extracted at each level. And in combination with key information such as the feature center and the closest clustering center, the noise attribute of each level is judged, and in combination with noise attribute judgment results of all levels, the overall noise attribute of the image is judged. The system and the equipment are used for executing the identification method. According to the method, the multi-level characteristic of the vehicle style label is combined, the noise conditions of different levels are fully considered, the noise of diffe