Abstract: Solar energy plays a crucial role in new power systems, making photovoltaic cell defect detection increasingly vital for transitioning to clean energy. Since traditional target detection models struggle to identify tiny defects on textured photovoltaic cells, this paper proposes an anthropomorphic vision biomimetic detection model. First, a backbone network inspired by human sensory fields and peripheral vision mechanisms is developed, incorporating an anthropomorphic visual attention mechanism and biomimetic feature extraction module to fully extract dynamic context while perceiving peripheral visual attention, correlating both to capture fine-grained defect features in noisy backgrounds. Second, for feature fusion, a separated spatial-semantic fusion pyramid is designed based on human brain information transfer patterns, with dedicated semantic and spatial information transfer modules in different pathways to enhance defective feature representation. Then, inspired by brain cortex partitioning mechanisms, an adaptively fused separated detection head is proposed to supervise multi-scale features adaptively while decoupling classification and localization tasks, with partitioned computation of location and category information. Finally, simulation experiments verify the model's effectiveness.