فهرست نشریات علمی دسترسی آزاد ایران

In this study, in the insurance industry, an integrated framework based on the Convolutional Neural Networks method is proposed to estimate the amount of car damage after the accident. This paper aims to analyze the performance of deep convolutional network methods in recognizing and classifying six different damage including surface damage, deep damage, car side mirror damage, glass damage, tire damage, and light damage. Regarding the case study, the required data for this research is obtained from the customers of an insurance company in 2021 and 2022 in Iran. This statistical population is a cross-section of all the pictures of accidental cars in the country and includes 20,000 pictures of people's cars after the accident, of which 4,100 pictures were selected and extracted as the target data of the research from the database of third-party insurance companies. The performance of models based on algorithms such as ConvNeXtBase, ConvNeXtXLarge, ResNet-50, ResNet-101, EfficienetNetB7, EfficienetNetV2L, and EfficientNetV2B0 has been evaluated under criteria including accuracy sensitivity, specificity, precision, and F- score criteria to compare all types of heuristic algorithms. The results of this research show that the performance speed of the models largely depends on the characteristics of the pre-trained models and the accuracy of the models is based on algorithms such as AlexNet, VGG-19, ResNet-50 have been obtained in the range of 0.59 to 0.91, which indicates the acceptable performance of these algorithms based on deep convolutional networks in detecting and evaluating car damage after accidents.