In recent years, generative adversarial and semi-supervised optimization of dynamic occlusion scenes has become a current research hotspot in agricultural automation. Generative Adversarial Network (GAN) and semi-supervised learning have demonstrated obvious value in solving the application and optimization strategy of agricultural image recognition, with a series of unresolved problems. This research review provides a detailed introduction to the current status of the research on generative adversarial and semi-supervised learning for dynamic occlusion scenes as well as the typical methods, with not only an efficient solution for the dynamic occlusion scene of mulberry leaves, but also important reference value in the development of agricultural intelligence and target detection tasks in similar scenarios. In future research, it is feasible to further deepen the cross-study of generative adversarial networks and semi-supervised learning, and promote theoretical innovation and practical application in related fields.