低信噪比下基于YOLOv3的昆虫目标检测

Insect Target Detection Based on YOLOv3 Under Low SNR

掌握昆虫迁飞规律对于农业防治和生态学研究具有重大意义,雷达正是检测昆虫迁飞最有效的手段。昆虫回波弱,传统的恒虚警检测(Constant False Alarm Rate,CFAR)算法在低信噪比(Signal To Noise Ratio,SNR)时的检测性能下降;同时昆虫目标体积小、飞行速度慢,在距离维和多普勒维的扩展性弱,特征少,在一维距离像上或者距离多普勒域基于深度学习的识别算法效果不佳。针对上述问题,本文提出了基于YOLOv3(You Only Look Once v3)网络的昆虫目标检测算法,通过短时傅里叶变换丰富目标的图像特征,利用图像特征对昆虫目标进行识别,提高了在低SNR下的检测率。进一步通过虚警-目标二元训练策略、目标检测置信度筛选策略降低了算法的虚警率。仿真和实测数据结果表明,所提算法在低SNR下的检测性能优于CA-CFAR算法,验证了算法的有效性。

Mastering the rules of insect migration was of great significance to agricultural control and ecological research.Radar was the most effective way to detect insect migration.Beacause of Insects’weak echoes,traditional Constant False Alarm Rate(CFAR)algorithm had poor detection performance under low signal-to-noise ratio(SNR).At the same time,because insect targets were small in size,slow in flight speed,weak in range and Doppler dimensions and showed few features,recognition algorithms based on deep learning in the One-dimensional distance profile or range Doppler domain did not work well.In response to the problems,this paper proposed a insect target detection algorithm based on YOLOv3,which enrich image features of target by short-time Fourier transform.Using image features to identify insect targets improved the detection rate under low SNR.Moreover,false alarm-target dual training strategy and target detection confidence selecting strategy was used to reduce false alarm rate.The results of simulation and measured data show that the detection performance of the proposed algorithm is better than CA-CFAR under low SNR,which verifies the effectiveness of the algorithm.