Volume 10,Issue 8
Research on Plateau Cattle and Sheep Detection Based on QHCSM-GAN Image Completion and Improved YOLO-OC Network
Highland animal husbandry is a core pillar of Qinghai Province’s agricultural and pastoral economy, and computer vision-based intelligent monitoring is a key approach to achieving refined and digital management of highland pastures. However, the complex environment at high altitudes presents challenges such as variable lighting, complex terrain, frequent target occlusion, and blurred imaging, leading to severe degradation in the quality of cattle and sheep images collected in the field and the loss of effective features. Furthermore, existing publicly available livestock datasets and general detection models have poor adaptability to open grazing scenarios on high plateaus, exhibiting shortcomings such as low detection accuracy and weak generalization ability. To address these issues, this paper proposes an integrated intelligent detection framework that combines image completion and target detection optimization. First, a dedicated Qinghai cattle and sheep dataset (QH-Livestock) covering various complex plateau scenes is constructed. Second, a generative adversarial network (QHCSM-GAN) for missing images of plateau cattle and sheep is designed, relying on a multi-path feature fusion encoder and a theoretically reliable discriminator to achieve degraded image restoration and missing feature completion. Finally, using YOLOv8n as the baseline model, a lightweight and high-precision YOLO-OC detection model is constructed by introducing full-dimensional dynamic convolution (ODConv) and convolutional block attention module (CBAM), enhancing multi-scale feature extraction capabilities and anti-interference capabilities against complex backgrounds. Experimental results show that QHCSM-GAN outperforms mainstream image completion algorithms in terms of structural similarity (SSIM), peak signal-to-noise ratio (PSNR), and learned perceptual patch similarity (LPIPS). The proposed YOLO-OC model achieves an accuracy of 84.6%, a recall of 81.2%, and an mAP50 of 82.5% on the QH-Livestock test set, with an inference speed of 109 FPS. This method effectively solves the core problems of poor image quality and insufficient detection accuracy of cattle and sheep targets in complex high-altitude scenes, and can provide reliable technical support for intelligent monitoring of high-altitude pastures and digital transformation of animal husbandry.
[1] Goodfellow I, Pouget-Abadie J, Mirza M, et al., 2014, Generative Adversarial Nets. Proceedings of the 28th International Conference on Neural Information Processing Systems, 2672–2680.
[2] Tian Y, Ye Q, Doermann D, 2025, YOLOv12: Attention-Centric Real-Time Object Detectors. arXiv.
[3] Li C, Zhou A, Yao A, 2022, Omni-Dimensional Dynamic Convolution. arXiv:2209.07947.
[4] Chen M, Zhou G, Ma Y, 2024, An Improved YOLOv8 Small Target Detection Algorithm With Embedded CBAM Attention. Computer Engineering and Science, 46(03): 478–485.
[5] Wang Y, Xu D, Wang Y, et al., 2025, TAF-ELAN: Tri-Attention Fusion Enhanced ELAN Network for Single Image Super-Resolution. Proceedings of the 2025 7th International Academic Exchange Conference on Science and Technology Innovation, 163–167.
[6] Liu S, Qi L, Qin H, et al., 2018, Path Aggregation Network for Instance Segmentation. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 8759–8768.
[7] Ma M, Liang L, Sheng X, 2025, MDF-Net: An Attention-Guided Multi-Scale Dual-Fusion Network for Retinal Vessel Segmentation. Measurement, 257(02): 118695.
[8] Choi Y, Choi M, Kim M, et al., 2018, StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation. Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 8789–8797.
[9] Isola P, Zhu J, Zhou T, et al., 2017, Image-to-Image Translation with Conditional Adversarial Networks. Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition, 5967–5976.
[10] Tang H, Xu K, Li X, et al., 2019, AttentionGAN: Unpaired Image-to-Image Translation Using Attention-Guided Generative Adversarial Networks. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 5647–5656.
[11] Jiang L, Li M, Lin L, et al., 2021, TSIT: A Simple and Versatile Framework for Image-to-Image Translation. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 423–432.
[12] Park T, Efros A, Zhang R, et al., 2020, Contrastive Learning for Unpaired Image-to-Image Translation. European Conference on Computer Vision, 319–334.
[13] Kim J, Kim M, Kim D, et al., 2019, CollaGAN: Collaborative Generative Adversarial Network for Missing Image Data Imputation. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 248–256.
[14] Zhou T, Li Z, Wang H, et al., 2020, Hi-Net: Hybrid-Fusion Network for Multi-Modal MR Image Synthesis. IEEE Transactions on Medical Imaging, 39(12): 4014–4025.
[15] Girshick R, 2015, Fast R-CNN. Proceedings of the IEEE International Conference on Computer Vision, 1440–1448.
