CN-121982372-A - Down on-line identification method based on multispectral microspur imaging and edge deep learning

CN121982372ACN 121982372 ACN121982372 ACN 121982372ACN-121982372-A

Abstract

The application is suitable for the technical field of product detection, and provides a down on-line identification method based on multispectral microspur imaging and edge deep learning, which comprises the steps of executing a vibration feeding instruction and a shaping and separating instruction, then obtaining the instant image information of the object to be detected, constructing multi-channel stack information according to the instant image information, inputting the multi-channel stack information to a convolution network to generate judging result information, and finally executing sorting instruction information based on the judging result information. The application can remarkably reduce the subjective difference and repeatability problems caused by manual microscopic experience, promote the consistency of cross batches, support batch identification and automatic recording under continuous sample supply conditions, and shorten release and sorting decision links. The method meets the real-time and stability requirements of industrial scenes, can also construct a traceable chain from the material to the output and is connected with the compliance marking under the existing standard system, and has better comprehensive cost performance in the large-scale material feeding inspection and sorting link.

Inventors

CHEN GUANJIE
YANG CHEN
Qi Zhengbao
ZHANG YU
Zha Shuangying
WANG YUJUE

Assignees

江西中纺联检验技术服务有限公司
江西服装学院

Dates

Publication Date: 20260505
Application Date: 20251226

Claims (9)

1. An on-line down feather identification method based on multispectral macro imaging and edge deep learning is characterized by comprising the following steps: Generating and executing a vibration feeding instruction based on a preset vibration feeding device, and generating and executing a shaping and separating instruction based on a preset airflow generating device; acquiring instant image information of an object to be detected based on preset shooting equipment and preset multispectral lighting equipment; constructing multi-channel stack information according to a plurality of the instant image information; Inputting the multi-channel stack information to a preset convolution network to generate judging result information, wherein the judging result comprises goose down type information or duck down type information; and generating and executing sorting instruction information based on the judging result information.
2. The method according to claim 1, wherein the multispectral lighting device comprises a plurality of preset lighting channel information, and the capturing device is used for acquiring the instantaneous image information of the object to be detected based on the preset shooting device and the preset multispectral lighting device: Generating and executing a first illumination instruction based on a preset multispectral illumination device, wherein the first illumination instruction is used for indicating the multispectral illumination device to illuminate based on first illumination channel information; responding to the first illumination instruction, and acquiring instant image information based on preset shooting equipment; generating and executing a second illumination instruction based on the multispectral illumination device after waiting for a specified time interval, wherein the second illumination instruction is used for indicating the multispectral illumination device to illuminate based on the next illumination channel information; acquiring next instantaneous image information based on the shooting equipment in response to the second illumination instruction; Circularly executing the second illumination instruction based on the multispectral illumination device after waiting for a specified time interval until a third illumination instruction is generated and executed, wherein the third illumination instruction is used for indicating the multispectral illumination device to illuminate based on the last second illumination channel information; generating and executing a fourth illumination instruction based on the multispectral illumination device after waiting for the time interval, wherein the fourth illumination instruction is used for instructing the multispectral illumination device to illuminate based on the last first illumination channel information; and responding to the fourth illumination instruction, and acquiring the last instant image information based on the shooting equipment.
3. The method of claim 1, wherein said constructing multi-channel stack information from a plurality of said transient image information comprises: Respectively carrying out target segmentation processing on each instant image information to generate preprocessed image information; And performing alignment processing on the plurality of preprocessed image information to generate the multi-channel stack information, wherein the multi-channel stack information comprises channel number information, image height information and image width information.
4. The method of claim 1, wherein the determination result information further comprises pending category information, wherein the inputting the multi-channel stack information to a predetermined convolutional network generates the determination result information, and wherein the generating comprises: Inputting the multi-channel stack information into a preset convolution network to generate the goose down type information and confidence information corresponding to the goose down type information; Or (b) Inputting the multi-channel stack information into a preset convolution network to generate the duck down type information and confidence information corresponding to the duck down type information; Or (b) Inputting the multi-channel stack information into a preset convolution network, and generating the undetermined category information and confidence information corresponding to the undetermined category information; The sorting instruction information includes first sorting instruction information, second sorting instruction information or skip instruction information, and the generating and executing sorting instruction information based on the determination result information includes: If the judging result information is goose down type information, generating and executing first sorting instruction information, wherein the first sorting instruction information is used for indicating starting a preset first air injection device to move an object to be detected to a preset goose down collecting box; if the judging result information is the duck down type information, generating and executing second sorting instruction information, wherein the second sorting instruction information is used for indicating to start a preset second air injection device so as to move an object to be detected to a preset duck down collecting box; If the judging result information is other category information, generating and executing the skip instruction information, wherein the skip instruction information is used for indicating a preset conveyor belt to move an object to be detected to a preset rechecking area; correspondingly, if the judging result information is pending type information, after the multi-channel stack information is input to a preset convolution network to generate judging result information, the method further comprises: and generating rechecking mark information based on the undetermined category information.
5. The method of claim 4, wherein after said generating and executing sort instruction information based on said determination result information, said method further comprises: uploading the instant image information, shooting time information corresponding to the instant image information, number information corresponding to the object to be detected, the judging result information and the confidence information to a designated storage database.
6. An on-line down feather recognition system based on multispectral macro imaging and edge deep learning, which is characterized by comprising: The vibration feeding instruction generation module is used for generating and executing a vibration feeding instruction based on a preset vibration feeding device and generating and executing a shaping and separating instruction based on a preset airflow generation device; The instantaneous image information acquisition module is used for acquiring instantaneous image information of an object to be detected based on preset shooting equipment and preset multispectral lighting equipment; The multi-channel stack information construction module is used for constructing multi-channel stack information according to a plurality of the instant image information; the judging result information generating module is used for inputting the multi-channel stack information to a preset convolution network to generate judging result information, wherein the judging result comprises goose down type information or duck down type information; And the sorting instruction information generation module is used for generating and executing sorting instruction information based on the judging result information.
7. The system of claim 6, wherein the system further comprises: And the instantaneous image information uploading module is used for uploading the instantaneous image information, shooting time information corresponding to the instantaneous image information, number information corresponding to the object to be detected, the judging result information and the confidence information to a designated storage database.
8. A terminal device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1 to 5 when the computer program is executed.
9. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the method according to any one of claims 1 to 5.

Description

Down on-line identification method based on multispectral microspur imaging and edge deep learning Technical Field The application relates to the technical field of product detection, in particular to a down on-line identification method based on multispectral microspur imaging and edge deep learning. Background Down (such as goose down or duck down) is used as a high-value filling material for clothing bedding, and the quality and tag compliance of the down directly affect the product performance and trade order. International and regional standards generally specify down/feather content, class, cleanliness, etc. from the point of view of "component labeling and label compliance". For example, european Union EN12934 requires marking the composition of 'down feather/feather' according to weight percentage and prescribing the labeling principle of waterfowl/land fowl sources, and GB/T17685 'down feather' and GB/T10288 'down feather test method' in China respectively standardize product classification and laboratory test methods so as to ensure the consistency of market identification and quality. The above criteria emphasize how to check and label, but do not provide a fast species identification technology route that can be "on-line, batch, objectified" in the production or circulation links. At present, the characteristics of the small branch knot form, pigment distribution, position and the like of the down feather subjected to manual sample separation are usually observed by a detector under a certain magnification to judge the down feather, the duck feather or the like, the microscopic examination mode is seriously dependent on the experience of manual sample separation and a detector, the down feather needs to be observed piece by piece, the detection quantity is limited and is difficult to match with the rapid judgment requirement of batch incoming materials, and the problem of poor reliability exists and needs to be further improved. Disclosure of Invention Based on the method, the embodiment of the application provides an on-line down identification method based on multispectral microspur imaging and edge deep learning, so as to solve the problem of poor reliability in the prior art. In a first aspect, an embodiment of the present application provides a method for on-line identification of down feather based on multispectral macro imaging and edge deep learning, where the method includes: Generating and executing a vibration feeding instruction based on a preset vibration feeding device, and generating and executing a shaping and separating instruction based on a preset airflow generating device; acquiring instant image information of an object to be detected based on preset shooting equipment and preset multispectral lighting equipment; constructing multi-channel stack information according to a plurality of the instant image information; Inputting the multi-channel stack information to a preset convolution network to generate judging result information, wherein the judging result comprises goose down type information or duck down type information; and generating and executing sorting instruction information based on the judging result information. Compared with the prior art, the method for identifying the down feather on line based on multispectral macro imaging and edge deep learning has the advantages that the terminal equipment can firstly generate and execute the vibration feeding instruction based on the vibration feeding equipment, and generate and execute the shaping separation instruction based on the airflow generating equipment, then quickly acquire the instantaneous image information of the object to be detected based on the shooting equipment and the multispectral lighting equipment, then effectively construct the multichannel stack information according to the instantaneous image information, then input the multichannel stack information into the convolution network, accurately generate the judging result information, and finally accurately generate and execute the sorting instruction information based on the judging result information, so that the quick judging requirement of accurately distinguishing the goose down feather and the duck down feather is realized, the reliability is improved, and the problem of poor current reliability is solved to a certain extent. In a second aspect, an embodiment of the present application provides an on-line down identification system based on multispectral macro imaging and edge deep learning, the system comprising: The vibration feeding instruction generation module is used for generating and executing a vibration feeding instruction based on a preset vibration feeding device and generating and executing a shaping and separating instruction based on a preset airflow generation device; The instantaneous image information acquisition module is used for acquiring instantaneous image information of an object to be detected based on preset shooting e