Image anomalies detection using transfer learning of ResNet-50 convolutional neural network

Abstract

With the quick advancement of keen fabricating, information-based blame determination has pulled in expanding attention. As one of the foremost prevalent strategies of diagnosing errors, deep learning has accomplished exceptional comes about. Be that as it may, due to the truth that the estimate of the seeded tests is little in diagnosing mistakes, the profundities of the deep learning (DL) models for fault conclusion are shallow compared to the convolutional neural network in other regions (including ImageNet), which limits the accuracy of the final prediction. In this paper, ResNet-50 with a 25 convolutional layer depth has been proposed to diagnose anomalous images. Trained ResNet-50 applies ImageNet as a feature extractor to diagnose errors. It was proposed on three sets of data which are the bottle, the spoon, and the carton, and the proposed method was achieved. The prediction accuracy of the data set was 99%, 95% and 90%, respectively.