Data Optimization Method for Ship Infrared Image Recognition

This study aims to address two challenging problems that affect the accurate and reliable recognition of ship infrared (IR) images in various scenarios: the interference of radiation highlights from specular reflections of target background sunlight and the difficulty of distinguishing ship targets from background confusion under low light conditions. We propose a data optimisation method for ship IR deep learning recognition, which uses processing tools such as median noise reduction and local histogram equalisation image enhancement to improve the performance and stability of ship IR recognition under the above unfavourable conditions. And the effectiveness of the method is verified on high fidelity mid-wave infrared (MWIR) ship simulation data using deep learning image classification algorithms such as AlexNet, VggNet and EfficientNet. To further demonstrate the effectiveness of the method, research has also adopted model feature gradient visualization technique, which indirectly verifies that our method can effectively improve the predictive reliability of ship recognition models. Overall, the data optimization method proposed in this study provides a feasible solution to address the challenges in ship IR image recognition and can provide a valuable reference for military mission-related research.