Development of a new model of step convolutional neural network for classification of anomalies on panoramas

A new model of a stepped convolutional neural network for classifying anomalies in panoramas has been developed. Appropriate datasets for classification are selected. The conclusion is made about the incompleteness of the method previously used by the authors to find anomalies in special areas with high color difference in panoramas. The search for these areas by the previously developed method did not set the task of their classification. For automatic identification of detected objects, it is proposed to apply deep learning models using suitable neural networks. Particular attention is paid to work with data containing unbalanced classes and images of different sizes. The results of image classification of popular architectures of neural networks are compared with the newly developed stepped convolutional neural network.