|
In complex sea environments, the signal-to-noise ratio and contrast of infrared images constantly change due to factors such as wave motion, wave reflection, changes in solar altitude angle, target distance, and uneven distribution of thermal radiation, resulting in missed and false detections in infrared ship detection. In the image, ship targets are generally present in the vicinity of the sea antenna. Therefore, by determining the position of the sea sky line, the approximate area of the ship target in the image can be determined, thereby suppressing possible interference from unrelated objects in the sky and sea outside this range and reducing the probability of false detection of infrared ship targets. And in order to solve the possible interference effects of sea clutter and clouds in traditional sea antenna extraction methods, this paper proposes a sea antenna difference extraction algorithm. By enhancing the gray vertical gradient difference between rows, the sea antenna is extracted. Compared with other typical sea antenna extraction algorithms, the accuracy of the sea antenna extraction method in this paper is improved by 7.2%. The infrared image of ships has significant noise, which is affected by the attenuation of atmospheric transmission overlong distances on the sea surface. The contrast of ship targets is low, making it difficult to directly extract ship features; Moreover, the infrared imaging of ship targets is greatly affected by factors such as season, weather, and lighting, especially in harsh weather conditions such as clouds, rain, and snow, which can seriously affect the imaging effect and make the infrared images of ships blurry. Therefore, in order to effectively achieve the detection and recognition of ship targets, this paper preprocesses the infrared images of ships before detection. An improved inverse harmonic filtering algorithm is proposed, which first adopts a gradient enhancement strategy, and then eliminates the bright spot interference of sea clutter through gamma transform and inverse harmonic filtering. This can effectively remove noise and enhance image contrast. In the infrared ship target detection algorithm method, this article improves the YOLOv4algorithm by adding a self attention module to enhance the network's detection and recognition ability for weak ship targets, and optimizes the neck structure to improve the network's recognition accuracy for ship targets of different scales. In addition, this article constructs an infrared ship image dataset generated through simulation and actual collection, which includes infrared ship images captured in different scenes, times, angles, and distances. By comparing the method proposed in this article with current mainstream detection methods through experiments, the actual detection accuracy is improved to over 90%.
|
