Visual Identification-Based Spark Recognition System

Abstract

With the development of artificial intelligence, image recognition has seen wider adoption. Here, a novel paradigm image recognition system is proposed for detection of fires owing to the compression of lithium-ion batteries at recycling facilities. The proposed system uses deep learning method. The SparkEye system is proposed, focusing on the early detection of fires as sparks, and is combined with a sprinkler system, to minimize fire-related losses at affected facilities. Approximately 30,000 images (resolution, 800 × 600 pixels) were used for training the system to >90% detection accuracy. To fulfil the demand for dust control at recycling facilities, air and frame camera protection methods were incorporated into the system. Based on the test data and realistic workplace feedback, the best placements of the SparkEye fire detectors were crushers, conveyors, and garbage pits.