The application of archive retrieval management system based on FCN and hash algorithm

With the rapid development of information technology, the demand for image feature extraction and efficient retrieval in archive management is constantly increasing. This study proposes an archive retrieval optimization method combining enhanced multi-scale dilated convolution with a fully convolutional network and a supervised contrastive learning hash network. The approach improves multi-scale feature extraction for archival images and generates efficient hash codes for fast retrieval. Experimental results show the model achieves an average feature extraction accuracy of 0.89 and a loss of 0.34 at 200 iterations. In retrieval tests, it demonstrates 78.9% accuracy and 326 ms delay in fuzzy distortion environments, with a feature encoding efficiency of 29.8MB/s for 10,000 samples. The method enhances feature extraction accuracy and retrieval efficiency, addressing the growing demand for intelligent and efficient archive management. It offers practical value for handling complex archival features and advancing archive management technology.