Abstract: This study attempts to use deep learning technology for the detection of the field cleanliness of third-generation low-light image intensifiers. By combining the Faster region-based convolutional neural network (RCNN) algorithm with the Residual Network (ResNet) and feature pyramid network (FPN) structures, the detection accuracy of field defects using the Faster RCNN algorithm was improved. The introduction of ResNet and FPN network structures addresses the limitations of the Faster RCNN network in complex scenes and the decline in multiscale target detection performance. The experimental results show that for the five typical defect types in the dataset, the improved Faster RCNN-ResNet-FPN network structure achieved a significant performance improvement in recognition accuracy. The mean average precision increased from 29.0% for the original Faster R-CNN to 72.2% for the proposed Faster R-CNN-ResNet-FPN model, demonstrating the effectiveness of the proposed method in detecting field defects of third-generation low-light image intensifiers and its potential applicability to similar inspection tasks. In summary, the third-generation low-light image intensifier field cleanliness detection method based on the Faster RCNN-ResNet-FPN algorithm achieved significant results, providing effective support for the expansion of third-generation low-light image intensifier detection technology in the future and laying a solid foundation for the development of subsequent field cleanliness detection technology.