Insect Target Detection Based on YOLOv3 Under Low SNR

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.