无人机目标检测的光适应预处理算法

Light Adaptive Preprocessing Algorithm for UAV Target Detection

无人机进行目标检测,检测精度受图像质量影响,图像质量又受光照条件的影响。因此,设计了一种基于分类的轻量级无人机光适应算法——Adaptive Enhance and Exposure (AEE),利用轻量级网络ShuffleNetv2对航拍图像按光照条件分类后,再利用改进的图像处理模块自适应光照转换器(Illumination Adaptive Transformer, IAT)进行图像增强或者曝光校正;为了解决航拍低光图像噪声干扰问题,设计边缘增强滤波模块。实验证明,轻量级网络ShuffleNetv2在自建航拍图像分类数据集上分类精度达到99.5%;加入边缘增强滤波模块的图像增强算法峰值信噪比(Peak Signal to Noise Ratio, PSNR)指标平均提升1.26,结构相似性指数(Structural Similarity Index, SSIM)指标提升0.023;改进后的IAT算法PSNR指标提升0.46,SSIM指标提升0.008;光适应算法对目标检测平均准确率提升0.4%,在Jetson AGX Xavier设备上处理速度达到57.4帧/秒,满足实时目标检测。

When UAVs perform target detection tasks,the detection accuracy is affected by image quality,and image quality is affected by lighting conditions.Therefore,a lightweight UAV light Adaptive Enhance and Exposure(AEE)algorithm based on classification is designed.The lightweight network ShuffleNetv2 is used to classify aerial images according to lighting conditions,and then the improved image processing module Illumination Adaptive Transformer(IAT)is used to perform image enhancement or exposure correction;In order to solve the problem of noise interference of low-light aerial images,an edge enhancement filtering module is designed.Experiments show that the classification accuracy on the self-built aerial image classification data set of the lightweight network ShuffleNetv2 reaches 99.5%;the Peak Signal to Noise Ratio(PSNR)of the image enhancement algorithm with the edge enhancement filtering module is improved by 1.26 on average,and a Structural Similarity Index(SSIM)is increased by 0.023;the PSNR index of improved IAT algorithm is improved by 0.46,and the SSIM index is increased by 0.008;the average accuracy of target detection by the light adaptation algorithm is increased by 0.4%,and the processing speed reaches 57.4 frames/s on the Jetson AGX Xavier device,which will not affect the real-time performance of the target detection algorithm.