Small part sorting method based on deep learning and sorting system thereof

Abstract

The invention discloses a small part sorting system based on deep learning and a sorting method thereof. The small part sorting system comprises a conveying belt, an optoelectronic switch, an industrial camera, a computer and a mechanical arm, the conveying belt is responsible for conveying parts, the optoelectronic switch is responsible for detecting whether the parts reach an image collecting area or not, the industrial camera is responsible for collecting images of the parts, and the computer processes the acquired images to obtain the types and the coordinate positions of the parts in theimages; and the mechanical arm completes grabbing and sorting of the parts with the determined types and positions. The sorting method comprises the following steps: S1, training a part identificationmodel based on a YOLOv4 target detection network, detecting the parts in a picture through the trained part identification model, and outputting type names and coordinate information of the parts inthe picture; S2, converting pixel coordinates of the part images into world coordinates; and S3, grabbing the parts into corresponding sorting boxes. According to the method and device, the coordinatepositions of the parts are obtained through model prediction, the part identification accuracy is higher, and the identification speed is higher.

Description

technical field [0001] The invention relates to a method for sorting small parts based on deep learning and a sorting system thereof, belonging to the field of industrial automation in intelligent manufacturing. Background technique [0002] With the advent of Industry 4.0 and the era of artificial intelligence, industrial production tasks are gradually handed over to robots, which greatly improves productivity while reducing costs, and at the same time liberates people from heavy, boring and extremely repetitive tasks come out. [0003] The sorting of parts on the early production line was mostly done by humans, which resulted in high labor costs and low efficiency, and manual sorting could not guarantee the efficiency and accuracy of sorting after long hours of work. The parts sorting of traditional machine vision is mainly based on artificially designed features, using methods such as feature points, minimum circumscribed rectangles and template matching to identify part...

AI Technical Summary

Problems solved by technology

This algorithm often cannot achieve satisfactory results for parts with different poses and complex shapes on the production line.

[0004] Aiming at the problems of weak adaptability to external environment changes, poor anti-disturbance ability and low recognition efficiency for complex and irregular parts when the traditional machine vision algorithm recognizes and sorts parts, the present invention selects a target detection method based on deep learning to complete each part on the production line. Identification of various parts, improving the accuracy of sorting

Images