Digital recognition of electric meter with deep learning

Abstract

Automatic meter reading (AMR) is faster and more efficient than manual meter reading by meter readers. Due to low cost and non-intrusiveness, image-based automatic meter reading attracts attention of researchers. However, existing image-based meter reading is only for specific type of electric meter, which is lack of adaptability. To deal with this problem, a method of meter reading based on deep learning that is applicable to different types of meters is proposed in this work. With the digital area recognition and meter type identification by deep learning and image enhancement, the effect of different appearances of digital screen different shooting angles and environmental interference can be eliminated, thus improving the adaptability of the method. Experimental results show that digital of different types of electric meters can be recognized efficiently by the proposed method.