Efficient power generation of photovoltaic cells plays a crucial role in promoting green and low-carbon circular development. To address the challenges posed by complex backgrounds and small target sizes in photovoltaic cell defect images, this paper proposes an improved defect detection model based on YOLO11n, named CCMW-YOLO11n. Firstly, a cross stage partial improvement (CSP-I) module is introduced into the backbone network of YOLO11n. This module integrates the multi-head self attention (MHSA), convolutional gated linear unit (CGLU), and conventional convolution (Conv), balancing global information perception and local feature extraction, thereby enhancing the extraction of multi-scale features. Secondly, the content-aware reassembly of features (CARAFE) upsampling technique is employed during the feature fusion stage. This method adaptively reorganizes feature maps and enhances details, effectively preserving fine-grained features and improving the model’s detection performance on complex targets. Additionally, the mixed aggregation net enhancement (MAN-E) module is incorporated into the neck network to further strengthen feature representation capabilities. Finally, addressing the limitations of the CIoU loss function in the baseline model, a novel bounding box regression loss function named Wise-Inner-SIoU is proposed by combining WIoUv3, Inner-IoU, and SIoU, optimizing the regression performance of the model. Experimental results demonstrate that the improved CCMW-YOLO11n model achieves a 9.6% increase in recall rate, with mAP@0.5 and mAP@0.5:0.95 reaching 91.0% and 61.1%, respectively, representing improvements of 3.1% and 2.0% over the baseline model, thereby realizing efficient detection of photovoltaic cell defects.