A deep learning based image recognition and processing model for electric equipment inspection

Electrical inspection is a daily significant check for electric utilities. Generally, electric utilities make inspection tour system and plans, assign employees to patrol insulators and transmission lines, collect faults or malfunction data and analyze it to assure normal state of electrical equipment. Obviously, the whole procedure is rather costly and time-consuming. In recent years artificial intelligence has arose and learning to automatically describe the content of images without human intervention becomes a research hotspot that explores the association between natural language process and computer vision. In this paper, we propose an image recognition and processing model applied to electrical inspection that analyzes various information sources such as time, position, geography and climate to help utilities with decision making as well as insulators and transmission line images gathered by patrol robots and unmanned aerial vehicles. The model is trained based on the recent advance in neural network that can be used to recognize and detect objects and then generate natural sentences describing an image. In other words, these sentences make up a summary of current condition and state of electrical equipment patrolled that is able to provide assistance, improving efficiency and cutting costs for inspection, operation and maintenance. The effect of our model is validated on our inspection image datasets.