CN-121988543-A - Sorting method and system based on visual AI in fabric sorting production line

CN121988543ACN 121988543 ACN121988543 ACN 121988543ACN-121988543-A

Abstract

The application relates to the field of article sorting, in particular to a sorting method and a sorting system based on vision AI in a fabric sorting production line, which comprise an image acquisition module for acquiring a visual angle image of the top of a fabric article on a conveyor belt, a feature extraction module for extracting multidimensional visual features and generating feature vectors based on a deep learning model processing image, a feature library building module for analyzing the feature vectors according to preset rules to identify paired articles and storing the paired articles in a correlated manner to build an updated feature library, a feature matching module for matching the feature vectors of the articles to be sorted with the feature vectors in the feature library in similarity and outputting a pairing judgment signal, a sorting execution module for executing sorting actions to sort the paired articles and single articles to different channels in response to the signals, and a control system for coordinating and controlling the whole sorting flow. The application achieves the technical effect of efficiently and accurately sorting paired articles and single articles in the fabric sorting production line.

Inventors

LI ZHENYU
HUANG ZHENG
Xu Kexu
LIN CHENG

Assignees

浙江联运知慧科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260121

Claims (10)

1. A vision AI-based sorting system in a fabric sorting line, comprising: The image acquisition module is used for acquiring top view images of fabric articles on the conveyor belt; The feature extraction module is in communication connection with the image acquisition module and is used for processing the image based on a deep learning model, extracting multi-dimensional visual features of the article and generating corresponding feature vectors; The feature library building module is in communication connection with the feature extraction module and is used for analyzing the extracted feature vectors according to a preset paired identification rule to identify paired objects and storing the two feature vector associations of the paired objects as feature vector pairs so as to build and update a feature library; the feature matching module is in communication connection with the feature extraction module and the feature library building module and is used for matching the similarity between the feature vector of the article to be sorted and the feature vector pair in the feature library and outputting a pairing judgment signal according to the matching result; A sorting execution module, which is in communication connection with the feature matching module, and is used for responding to the pairing judgment signal and executing sorting action to sort the paired articles and the single articles to different output channels; and the control system is respectively in communication connection with the modules and is used for coordinating and controlling the whole sorting flow.
2. The vision AI-based sorting system of claim 1, wherein the image acquisition module includes a high resolution industrial camera fixedly disposed above the conveyor belt at a top view angle and an annular light supplementing device providing adjustable illumination for the industrial camera.
3. The vision AI-based sorting system of claim 1, wherein the deep learning model employed by the feature extraction module is a deep convolutional neural network, the network structure comprising, in order: an input layer for receiving the preprocessed image data; A multi-stage convolution layer for extracting feature maps of different levels from the image data; The feature fusion layer is used for fusing texture features, pattern features and shape features from the multi-stage convolution layer; And the full-connection output layer is used for mapping the fused features into high-dimensional feature vectors.
4. The vision AI-based sorting system in a fabric sorting line of claim 1, wherein the preset pairing recognition rules include at least one of shape symmetry analysis rules and texture matching calculation rules.
5. The vision AI-based sorting system of claim 1, wherein the feature matching module measures similarity by calculating euclidean distance between feature vectors of the items to be sorted and pairs of feature vectors in a feature library and compares the distance to a preset threshold to generate the pairing decision signal.
6. The vision AI-based sorting system of claim 1, wherein the sort execution module includes a robotic arm, a flexible gripping device mounted at a distal end of the robotic arm, and a dual-channel sorting conveyor belt including at least one paired article channel and one single article channel.
7. A visual AI-based sorting method in a fabric sorting line, applied to a visual AI-based sorting system in a fabric sorting line according to any one of claims 1 to 6, comprising the steps of: S1, acquiring an image of a fabric article to be sorted through an image acquisition module; S2, extracting visual features of the object and generating feature vectors of the visual features through the feature extraction module after preprocessing the image; S3, feature library construction, namely identifying paired objects according to preset rules by the feature library construction module in a system initialization or learning stage, and storing feature vectors of the paired objects as feature vector pairs into a feature library; S4, in a real-time matching step, during sorting operation, matching the similarity between the feature vector of the article to be sorted and the feature vector pair in the feature library through the feature matching module to obtain a pairing judgment result; S5, sorting execution step, namely driving the sorting execution module to sort the articles to the corresponding output channels according to the pairing judgment result.
8. The vision AI-based sorting method in a textile sorting line of claim 7, wherein in the feature extraction step, the preprocessing operation on the image includes at least one of image denoising, illumination compensation, background segmentation, and contour extraction.
9. The visual AI-based sorting method in a fabric sorting line of claim 7, wherein in the feature extraction step, the extracted and fused visual features include texture features, pattern features, brand identification features, and shape features.
10. The method of vision AI-based sorting in a fabric sorting line according to claim 7, characterized in that the sorting performing step comprises in particular: If the pairing judgment result is paired objects, controlling the sorting execution module to transfer the paired objects to the paired object channels; and if the pairing judgment result is a single article, controlling the sorting execution module to transfer the single article to the single article channel.

Description

Sorting method and system based on visual AI in fabric sorting production line Technical Field The application relates to the field of article sorting, in particular to a sorting method and a sorting system based on vision AI in a fabric sorting production line. Background In the technical field of intelligent sorting and machine vision, along with the continuous increase of waste clothes and textile recovery processing demands, the method is of great importance to the development of efficient and accurate sorting technology of fabric articles. The technical progress in the field can obviously improve the efficiency and quality of recovery treatment and promote the development of the textile recovery industry to an intelligent direction. The efficient sorting technology can reduce labor cost, improve resource utilization rate and positively influence environmental protection and sustainable development. In the development of related industries, how to realize accurate identification and classification of fabric articles becomes a key problem. In the prior art, in the waste clothes and textile recovery treatment scene, two modes are mainly adopted in the traditional way in the face of the situation that a large number of scattered single pieces and paired fabric articles are mixed. One is a manual sorting method, in which a worker sorts fabric articles by his own experience and visual inspection. Another existing automated sorting solution is mostly focused on identifying fabric materials, such as distinguishing different materials of cotton, polyester, etc., or classifying fabrics by color. However, the conventional manual sorting method has significant drawbacks in that it is inefficient, requires a lot of manpower and time costs, and in long-time work, the manual sorting is prone to erroneous judgment and omission. However, the existing automatic sorting solution, although having a certain effect in terms of material identification and color classification, lacks the capability of accurately identifying the characteristics of paired fabric articles, and generally does not support real-time matching and classification in a complex environment, so that the problems of article type change, placement angle and the like are difficult to deal with. Disclosure of Invention The application aims to solve the problems that the traditional manual sorting efficiency is low and the existing automatic sorting system lacks the capability of identifying articles, and provides a sorting method and a sorting system based on visual AI in a fabric sorting production line, so that the recycling efficiency and quality of waste fabrics are improved. In order to achieve the above purpose, the following technical scheme is provided: A vision AI-based sorting system in a fabric sorting line, comprising: The image acquisition module is used for acquiring top view images of fabric articles on the conveyor belt; The feature extraction module is in communication connection with the image acquisition module and is used for processing the image based on a deep learning model, extracting multi-dimensional visual features of the article and generating corresponding feature vectors; The feature library building module is in communication connection with the feature extraction module and is used for analyzing the extracted feature vectors according to a preset paired identification rule to identify paired objects and storing the two feature vector associations of the paired objects as feature vector pairs so as to build and update a feature library; the feature matching module is in communication connection with the feature extraction module and the feature library building module and is used for matching the similarity between the feature vector of the article to be sorted and the feature vector pair in the feature library and outputting a pairing judgment signal according to the matching result; A sorting execution module, which is in communication connection with the feature matching module, and is used for responding to the pairing judgment signal and executing sorting action to sort the paired articles and the single articles to different output channels; and the control system is respectively in communication connection with the modules and is used for coordinating and controlling the whole sorting flow. The technical scheme is adopted, an image acquisition module is utilized to acquire a visual angle image of the top of a fabric article, a data basis can be provided for subsequent analysis, a feature extraction module processes the image based on a deep learning model and extracts multidimensional visual features to generate feature vectors, article features can be accurately acquired, a feature library building module identifies paired articles and stores feature vector pairs according to preset rules, a feature library is built and updated to provide data support for matching, a feature matching module performs similarity matching