Infrared aircraft target recognition technologies under complex interference conditions have been a research hotspot. The development of infrared countermeasure technologies has made the air combat environment increasingly polymorphic, thus hindering the accurate recognition of infrared targets. In particular, a complex artificial interference can severely obscure target, resulting in the loss of continuity and saliency of target features, and the inability to fully describe the characteristics of a target object, thus making aerial target recognition inaccurate. To overcome this problem, we propose an anti-interference recognition algorithm for air targets based on mixed depth features. First, to improve the ability of the algorithm accurately in selecting candidate regions in a full trajectory, an improved fuzzy C-means (FCM) adaptive segmentation algorithm is developed to extract the target candidate region. Second, a deep convolutional feature (convolutional neural network) and the histogram of gradient feature are used to construct hybrid depth features for a more comprehensive representation of the target. Finally, to consider the diverse confrontation scenarios between the target and the interference in a combat scene, a three-class support vector machine model is proposed to deal with the target, interference, and interference adhesion states. The experimental results show that in a complex interference environment, the air target anti-interference recognition algorithm based on the mixed depth features exhibits an accuracy rate of 92.29%, which represents an accurate target anti-interference recognition.
Read full abstract