Streetlight Working or Not-Working Classification

About Model

This AI model performs automated visual inspection of streetlights using a MobileNetV2-based deep learning image classification architecture built using the Aargus DIY visual inspection tool. It classifies input images into two categories: Working and Not Working. The model uses ImageNet-pretrained MobileNetV2 with transfer learning and selective fine-tuning. A custom classification head consisting of Global Average Pooling, Batch Normalization, Dense layers and Dropout is used for binary classification. Data augmentation and class weighting are incorporated during training to improve generalization and address class imbalance. The model is intended for municipal and utility infrastructure monitoring, smart-city applications, automated streetlight inspection and maintenance prioritization. It can be applied to individual field images or video frames extracted from inspection footage. Reported evaluation performance is 95.16% accuracy, 91.30% precision and 95.45% recall. Actual performance may vary with image quality, lighting, camera viewpoint, environmental conditions and differences between deployment data and training data. The model provides visual inspection support.