Infrared Video Data Cleaning Based on LightWeight CNN-LSTM

Video cleaning is a pivotal technology for enhancing the quality of infrared video data. In order to achieve high-quality cleaning of infrared video data, the temporal correlation features in the video data are thoroughly taken into account. A feature extraction model is designed. This model integrates Convolutional Neural Networks (CNN) and Long Short-Term Memory Networks (LSTM). It can extract infrared video features in both spatial dimension and temporal dimension. Considering the characteristics of low resolution and weak feature information in infrared video images, a lightweight strategy is adopted to achieve high-quality extraction of weak information in infrared images. In this paper, an infrared video data cleaning model lightweight CNN-LSTM is developed by incorporating a lightweight strategy and video spatiotemporal features. Through seven experimental comparisons, the model utilized in this study achieved improvement ranges of 0.6% to 43.8% in accuracy and 0.7% to 65.7% in recall.