Lightweight cruise missile object tracking algorithm based on feature fusion
Aiming at the situation where the visual target tracking algorithm on the cruise missile platform began to incorporate deep learning techniques but was limited by the computational power of the hardware platform,a lightweight deep learning-based Siamese network framework was proposed.This framework ensured the tracking performance of the cruise missile while operating under lower computational load.Based on an IOU quality evaluation branch and a regression branch,a new feature fusion method was proposed.By using 1×1 convolution to adjust the number of channels in the feature layers and controlling the ratio of channel numbers,the channels of different feature layers were concatenated proportionally according to their depths and used for subsequent feature fusion modules.Before feature concatenation,a method called feature combination was employed to obtain fused features with varying receptive fields,thereby further enhancing feature resolution.The proposed new feature fusion method and feature combination approach combine features vertically and horizontally,effectively utilized the feature attributes and improved algorithm performance.According to the hardware platform attributes of the cruise missile,a lightweight backbone network called AlexNet was adopted.The algorithm was tested on three datasets:OTB100,GOT-10K,and UAV123,the overall structure ensures a high accuracy and success rate at 160 fps real-time running speed.On the basis of satisfying the special working environment of cruise missiles,advanced relatively tracking performance has been achieved.The overall framework is relatively simple yet excellent,demonstrating good real-time tracking capabilities.Other modules can be integrated to further enhance the tracking performance.
NETL
NSTL
万方数据
