Power image retrieval based on adaptive weight and multi-dimensional image feature fusion

Large amounts of unstructured data are generated during the process of power production. However, due to their complex nature and difficulties in analysis, their value is not effectively utilized. Effective power image feature extraction and accurate retrieval are important to efficiently utilize power images. Accordingly, this study proposes a multi-dimensional feature fusion-based power image retrieval method. This approach improves the way image features are expressed by leveraging weight calculations and image feature fusion. The first step is using three different feature extraction models separately to extract power image features and perform image retrieval. Next, we use an objective evaluation method to calculate the weights of the features based on the accuracy of the retrieval results. This helps in avoiding subjectivity and one-sidedness. Once we calculate the feature weights, we combine the features extracted by the three models and use the fused features for image retrieval. Experimental results demonstrate that our proposed method has an average retrieval accuracy of 91.5% and outperforms other similar methods in retrieval accuracy and can better support power production.