The work aims to solve the problems of low efficiency and poor detection accuracy in the existing sorting and detection methods of mixed color in industrial wires. An wire sorting device based on machine vision technology and an sorting detection approach for mixed-color wires in combination with image processing techniques and deep learning principles are proposed. Firstly, the region of interest in the image was selected manually to segment images of wire connector and wires, and methods of template matching and color localization were applied to realize the identification of the connector's front and back sides and monochrome wires. Secondly, the PE mixed-color wire datasets were produced, and the detection effect of four object detection algorithms, i.e., Faster R-CNN, SSD, YOLOv3, and YOLOv5m was researched. The experimental results indicate that the YOLOv5m model possesses the best capacity between the detection speed and accuracy, which decreases the detection time by 18.55% with an average recognition accuracy of 98.83%, which can be applied to the sorting detection of a variety of industrial wires.