CN-122023463-A - Visual sorting method and system for pollutants in insulating material particles

CN122023463ACN 122023463 ACN122023463 ACN 122023463ACN-122023463-A

Abstract

The invention belongs to the technical field of image processing, and particularly relates to a visual sorting method and a visual sorting system for pollutants in insulating material particles, wherein the method comprises the steps of obtaining particle characterization size, adjacent particle number and adjacent particle distance; the method comprises the steps of obtaining a target crowding degree, calculating a target deviation index used for representing the moving environment of target particles, obtaining a real-time moving distance according to the position change before and after the movement of the target particles, obtaining a total moving distance from a database, obtaining target dynamic noise, outputting target final coordinates and eliminating, judging whether eliminating is successful or not and storing structured data, and optimizing a motion prediction model. The invention solves the technical problems that the traditional sorting mode is inaccurate in particle motion trail prediction and cannot adapt to complex environmental changes.

Inventors

GUO XIONGTAO
YAO WANG
WANG ZHANSHENG
LIU BINGJI

Assignees

淳化昆仑优佳电缆有限公司

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. A method of visually sorting contaminants in particles of insulating material, comprising: the method comprises the steps of firstly positioning target particles based on a neural network model, obtaining particle characterization size, the number of adjacent particles and the distance between the adjacent particles according to morphological characteristics and distribution of the target particles and the adjacent particles, accumulating crowding contribution caused by single adjacent particles obtained by the size and distance of the adjacent particles in a preset adjacent region of the target particles, and obtaining target crowding degree; Calculating a target deviation index for representing the moving environment of the target particles based on the regulating action of the deviation degree of the target crowding degree relative to the historical condition according to the fact that whether the historical working condition is stable or not; Obtaining a real-time moving distance according to the position change before and after the movement of the target particles, and obtaining a total moving distance from a database; updating the motion prediction model based on the target dynamic noise, outputting the final coordinates of the target, removing, judging whether the removing is successful or not, storing the structured data, carrying out regression analysis on a plurality of failure data in the structured data, and optimizing the motion prediction model.
2. A method of visual sorting contaminants in particles of insulating material according to claim 1, characterized in that said obtaining a particle characterization size, a number of adjacent particles, a distance between adjacent particles, comprises: Based on visual characteristics such as material particle color, texture, morphology and the like in real-time materials, pollutant particles are accurately identified and segmented from a normal insulating material particle background and are marked as target particles, the particle characterization size is obtained by calculating the number of pixels contained in a segmentation mask of the material particles, the pixel coordinates of the target particles are output, all adjacent particles are identified in a preset neighborhood of the target particles by using an image processing algorithm, the number of the adjacent particles is obtained, the pixel distances between all the adjacent particles and the target particles are calculated, the conversion is carried out according to camera calibration parameters, and the pixel distance between each adjacent particle and the target particle is marked as the adjacent particle distance.
3. A method of visual sorting contaminants in particles of insulating material according to claim 1, characterized in that said target crowding degree satisfies the following expression: ; In the formula, Representing a target congestion degree; representing the first of the target particles Particle characterization size of individual adjacent particles; Representing the target particle and the first Adjacent particle distance of adjacent particles; representing a previously obtained distance decay index; representing the number of adjacent particles of the target particle; Representing an extremely small positive number, and ensuring that the denominator is not 0; Representing the normalization function.
4. A method of visual sorting contaminants in particles of insulating material according to claim 1, characterized in that said target deviation index satisfies the following expression: ; In the formula, Representing a target deviation index; Representing a target congestion degree; Representing a historical average congestion level; Representing the standard deviation of the historical congestion degree; 、 Representing an extremely small positive number, and ensuring that the denominator is not 0; as a hyperbolic tangent function.
5. A method for visual sorting of contaminants in particles of insulating material according to claim 1, wherein said obtaining a real-time movement distance from a change in position of a target particle before and after movement, obtaining a total movement distance from a database, comprises: The pixel sitting of the target particles obtained at the first time is marked as primary target coordinates, real-time position coordinates are obtained according to the optical flow positioning of the target particles in the moving process of the target particles, euclidean distance between the primary target coordinates and the real-time position coordinates is calculated, real-time moving distance is obtained, and the inherent distance of the system is obtained from a database and is recorded as total moving distance.
6. A method of visual sorting contaminants in particles of insulating material according to claim 1, characterized in that said target dynamic noise satisfies the following expression: ; In the formula, Dynamic noise is the target; is a target deviation index; Representing a historical average congestion level; Representing the standard deviation of the historical congestion degree; is a baseline process noise; representing a real-time movement distance; representing the total movement distance; Representing an absolute value function; Representing the minimum positive number, ensuring that the denominator is not 0.
7. A method of visual sorting contaminants in particles of insulating material according to claim 1, wherein said outputting target final coordinates and rejecting includes: The system updates the motion prediction model based on target dynamic noise, calculates a target final coordinate when target particles reach the rejection execution module at a certain prediction moment, the target final coordinate is sent to the rejection execution module, and meanwhile, an air tap accurately corresponding to the target final coordinate is instantaneously activated to jet air flow, so that the target particles are accurately separated from the main material flow.
8. A method of visual sorting contaminants in particles of insulating material according to claim 1, wherein said determining whether culling is successful and storing structured data includes: The method comprises the steps of setting a verification camera at the downstream of a rejection execution module, wherein the camera is used for capturing a material flow image of a rejection area just passed, analyzing the image captured by the verification camera by a system to judge whether target particles are completely separated from a main material flow and no new pollutant particles appear in the image, judging that the rejection is successful if the target particles are completely separated and no new pollutant exists, judging that the rejection is failed if the target particles are still in the main material flow or the new pollutant appears, and marking the verification result of the rejection task and the whole set of parameters including the target crowding degree, the target deviation index, the target dynamic noise and the target final coordinate in the rejection task process as structured data.
9. A method of visual sorting contaminants in particles of insulating material according to claim 1, characterized in that said optimizing a motion prediction model comprises: The system periodically counts the structured data, analyzes the failure data with failure labels in all the structured data, carries out regression analysis on track data of a plurality of failure data with failure labels in all the structured data, updates the target dynamic noise in the motion prediction model, and completes the optimization of the motion prediction model.
10. A visual sorting system of contaminants in particles of insulating material, comprising a processor and a memory, the memory storing computer program instructions which, when executed by the processor, implement a method of visual sorting of contaminants in particles of insulating material according to any one of claims 1 to 9.

Description

Visual sorting method and system for pollutants in insulating material particles Technical Field The invention relates to the technical field of image processing. More particularly, the present invention relates to a method and system for visual sorting of contaminants in particles of insulating material. Background In the production process of high-performance insulating materials, the removal of trace pollutants in raw materials is a core link for guaranteeing the quality and performance of final products, so that a color sorter or sorting system based on machine vision is commonly adopted in the industry, falling material flows are captured through a camera, pollutant particles with abnormal colors or morphologies are identified, the movement track of the pollutant particles is calculated by a prediction system, and finally, a high-pressure air nozzle is driven to spray and remove the pollutant particles at accurate time and position. However, most of the conventional visual sorting systems are passive prediction, that is, assuming that the particles do ideal free fall or parabolic motion under the action of gravity, it is difficult to capture abnormal states in the prediction process, such as trajectory deviation of the particles caused by collision with the accompanying particles or disturbance of uneven airflow during the falling process. These random disturbances, which are not considered by the model, although slightly affected in a single event, accumulate into significant prediction errors in large-scale, high-density production streams, eventually leading to inaccurate rejection and hidden product quality hazards. In order to overcome the defect, some methods for early warning of abnormality by introducing advanced prediction algorithms such as Kalman filtering and the like appear in the prior art, namely, the particle positions acquired in real time are compared and corrected with a dynamic model based on fixed parameters. However, this determination method based on the fixed parameter model has an inherent limitation that in a real industrial site, the disturbance intensity of the particles is not constant, but is affected by various dynamic factors such as the material flow density, the electrostatic interaction among the particles, the environmental humidity and the like. The dynamic changes cannot be accommodated by using a single, fixed model parameter, such as a fixed process noise covariance. Under the working conditions that material flows become crowded and frequently collide, the dependence of a system on a model prediction result is too high, the correction of the actually observed position is insufficient, missing report is generated, namely pollutants are not removed, under the working conditions that the material flows are sparse and the flight track is stable, the prediction track is excessively oscillated due to the fact that model parameters are too conservative, false report is generated, namely good products are removed by mistake, and the lack of the capability of sensing and self-adaptive adjustment on the current real flight environment of particles makes the prior art difficult to balance between high removal precision and high material yield, and severely restricts the reliability and practicability of the intelligent visual sorting technology. Disclosure of Invention In order to solve the technical problems that the particle motion trail is inaccurate to predict and cannot adapt to complex environmental changes in the traditional sorting mode, the invention provides the following aspects. In a first aspect, the present invention provides a method of visually sorting contaminants in particles of insulating material, comprising: The method comprises the steps of firstly positioning target particles based on a neural network model, obtaining particle characterization sizes, the number of adjacent particles and the distance between the adjacent particles according to morphological characteristics and distribution of the target particles and the adjacent particles, accumulating crowding contributions caused by single adjacent particles obtained through the sizes and the distances of the adjacent particles in a preset adjacent region of the target particles, obtaining target crowding degree, calculating target deviation indexes for representing the moving environment of the target particles according to the adjustment effect of deviation degree of the target crowding degree relative to historical conditions according to whether historical working conditions are stable or not, obtaining real-time moving distances according to position changes before and after the movement of the target particles, obtaining total moving distances from a database, carrying out self-adaptive adjustment on pre-obtained reference process noise through fusion of the real-time moving progress, the historical working conditions and the target deviation indexes of the target particles, obtaining target