An Improved Dual-Path Attention Mechanism With Enhanced Discriminative Power for Asset Comparison Based on ResNet50

With the wide application of deep convolutional neural networks (CNNs), higher requirements are put forward for feature extraction and discriminant ability in asset comparison tasks. The rapid development of CNNs has significantly promoted the performance improvement of computer vision tasks, in which ResNet50, as the mainstream feature extraction network, effectively alleviates the problem of gradient disappearance in deep networks by means of residual learning. However, despite ResNet50's excellent performance in many visual tasks, there is still room for improvement in the effectiveness of feature expression and the mechanism of attention to key areas in response to changes in complex scenes. Therefore, this paper puts forth a new network architecture with improved dual-path attention mechanism and discriminative feature enhancement module based on ResNet50 (DADF-ResNet50). This method strengthens important features in channel and spatial dimensions respectively through the improved dual-path attention mechanism to improve the model's focusing ability on key asset areas. At the same time, the discriminating power enhancement module further optimizes the distinguishing ability of features and effectively improves the feature comparison performance between different assets. We design and implement the network architecture of this approach and evaluate it comprehensively in the asset comparison task. The experimental results show that the model has better distinguishing ability and feature extraction efficiency in diverse asset comparison scenarios, which is helpful to solve the asset identification problem in complex scenarios.