Unmanned aerial vehicle camera target detection method based on image processing and YOLOv5

Existing unmanned aerial vehicle camera target detection methods suffer from insufficient accuracy and low efficiency. To address this situation, this study proposes an unmanned aerial vehicle camera target detection method based on image processing and YOLOv5. The study first introduces recursive network and coding and decoding structure to optimize the generative adversarial network, and applies it to the deblurring process of unmanned aerial vehicle images. After that, luminance enhancement network is utilized for further enhancement of images with low luminance condition. The processed image is output to the improved you look only once version 5 model for target detection. The outcomes indicated that the structural similarity, peak signal-to-noise ratio, and brightness improvement contrast coefficient of the unmanned aerial vehicle aerial images processed by the proposed method were 0.94, 29.53dB, and 1.75, respectively. The detection accuracy of the target detection method was over 90% for all the different camera scenes. Its mean average precision was 0.94 and frames per second was 25.36 frames/s. The proposed target detection method can effectively improve the image quality and extract effective information with high precision. This provides reliable data guarantee for unmanned aerial vehicle work.