TinyML Autoencoder for Transmission Line Anomaly Detection

Power distribution infrastructure is vital for reliable energy delivery but faces challenges from environmental stressors, aging components, and inefficient inspections. This paper introduces an automated anomaly detection system using image processing and deep learning to enhance transmission line inspections. The framework employs a convolutional autoencoder for real-time defect identification, detecting subtle degradation patterns missed by manual methods. A key innovation is the deployment of the optimized model on Raspberry Pi, reducing its size from 306KB to 2KB (a 99.3% reduction) without significant performance loss, enabling efficient edge computing. Experimental results show improved detection accuracy, reduced inspection time, and lower costs. By enabling predictive maintenance, the system enhances grid reliability, worker safety, and operational efficiency. This work demonstrates the potential of lightweight deep learning models in modernizing electrical infrastructure inspections and lays the groundwork for future resource-constrained automated systems.