Based on the improvement of YOLO v3 aluminum alloy door and window profile identification method

Abstract

In the traditional aluminum alloy door and window profile production process, the identification of profiles mainly rely on RFID electronic labels and staff eye identification, but aluminum alloy doors and windows profiles are a wide variety, while some profile models are similar, staff difficult to judge, RFID electronic labels also have recovery difficulties, higher cost problems. Based on this, this paper proposes an aluminum alloy door and window profile identification method based on machine vision and deep learning, with YOLO v3 as the detection network, adding a separable convex network to the Darknet neural network, improving the detection accuracy and speed, and improving the K-mean clustering method to obtain a more accurate prediction box. The experimental results show that compared with SSD network and YOLO v3 network, the accuracy of improved detection network recognition is up to 0.15s, which can meet the needs of industrial production.