Due to the shallow depth of field inherent in optical microscopic imaging of micron-scale spores, defocused blurred images frequently occur, resulting in the loss of high-frequency information and blurred edges of urediniospores. Defocus deblurring restoration is crucial for the subsequent accurate detection of spore targets, making it a critical technical foundation for the early prevention and control of airborne wheat stripe rust. To address the issues of high-frequency information loss and blurred edges of urediniospores caused by defocused blurring in microscopic imaging of spore capturing devices under complex conditions, an improved DeblurGANv2 method for deblurring microscopic images of urediniospores was proposed. Firstly, a bottom-up five-level feature enhancement module after the feature fusion module of DeblurGANv2 was introduced, which shortened the propagation path from shallow features to deep features, enhancing the mutual fusion of features at different scales and improving the model’s ability to restore high-frequency and spore edge features. Simultaneously, the convolutional block attention module (CBAM) was incorporated into the feature extraction backbone network, increasing the weight of urediniospore feature information in both spatial and channel dimensions, enhancing the model’s capacity to express urediniospore features and enriching the detail information in the restored images. Finally, four mainstream object detection models, including YOLO v5, Faster-R CNN, CenterNet, and YOLO v8 were employed to detect urediniospores in images before and after restoration, comparing the impact of the improved DeblurGANv2 model on detection performance. The experimental results indicated that the improved DeblurGANv2 restoration model achieved mean squared error (MSE), peak signal-to-noise ratio (PSNR), and structural similarity (SSIM) values of 0.001 4, 28.88 dB, and 0.966, respectively, representing improvements of 17.65%, 3.29%, and 0.35% over the original DeblurGANv2 model, respectively. Furthermore, the four object detection models exhibited varying degrees of performance enhancement when combined with the improved DeblurGANv2 model. Among them, the YOLO v8 model, utilizing the improved DeblurGANv2 restoration, demonstrated the best performance, with precision (P), recall (R), and mean average precision (mAP@0.5) values of 96.1%, 95.1%, and 97.7%, respectively, an increase of 3.0, 5.0, and 23.6 percentage points compared with that using the YOLO v8 detection model directly. This validated the effectiveness of the proposed improved DeblurGANv2 model in recovering defocused and blurred spore information in microscopic images, significantly enhancing the detection performance of spore detection models and providing technical support for the airborne detection of urediniospores.