基于改进DeblurGAN-v2的柑橘病虫害图像去模糊算法OA

Aiming atthe issue of image blur encountered during real-time acquisition and detection of citrus pest and di-sease images,which originated from motion blur induced by unmanned aerial vehicle(UAV)movement and out-of-focus blur resulting from imprecise camera focusing.An efficient deblurring algorithm method was proposed,that is,a deblur-ring preprocessing step was added before the target detection algorithm to improve the image clarity and enhance the detec-tion accuracy and robustness.To achieve lightweight and efficient deblurring,this study employed a FPN-MobileNetv3-small lightweight architecture within the backbone of the DeblurGAN-v2 model.Moreover,the Selective Kernel Networks(SKNet)attention mechanism was introduced to enable adaptive selection of convolutional kernel sizes,thereby enhancing algorithmic robustness.Furthermore,Self-Calibrated Convolutions was leveraged to dynamically adjust the receptive field of each convolution in the intermediate layers,enriching feature representation,which actually solved the problem that the details were easy to be lost and the feature fusion effect was not ideal in the deblurring process.Experimental results dem-onstrated that,compared with the original model,the improved model achieved a peak signal-to-noise ratio(PSNR)in-crease of 3.25 dB,a structural similarity index(SSIM)increase of 9.26%,a model size of 16.4 M,and a processing speed of 41.7 FPS.By utilizing a YOLOv8 model for object detection on UAV-captured orchard citrus pest and disease images,the results indicated that with no significant reduction in the model recall rate,the precision(P)and mean average precision(mAP)of detection were improved by 3.8%and 1.8 percentage points,respectively,thus validating the efficacy of the proposed deblurring algorithm.This research provided higher-quality images for citrus pest and disease detection,thereby contributing significantly to the realization of precision agriculture and the enhancement of the economic value of agricultural products.