Online monitoring and sorting system for quality of flame retardant particles based on AI visual recognition

Abstract

The present application relates to the technical field of flame retardant particle quality monitoring, in particular to an online monitoring and sorting system for the quality of flame retardant particles based on AI visual recognition, which comprises a polarized light field imaging module, a refractive index gradient field construction module, a dynamic sorting decision module and an online self-optimization module. The polarized light field imaging module is used to obtain the Stokes parameter image and the multispectral transmittance distribution map of the particles; the refractive index gradient field construction module is used to convert the optical data into a refractive index gradient field, accurately identify internal defects; the dynamic sorting decision module is used to generate sorting instructions according to the defect characteristics; and the online self-optimization module is used to realize continuous optimization of the system. The present application solves the problem that the traditional method is difficult to accurately identify transparent defects, can effectively identify micron-level pure bubbles and transparent cracks, realizes defect classification and cause analysis, and can meet the real-time quality monitoring requirements of the flame retardant production line, and provides an intelligent solution for product quality control.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of flame retardant particle quality monitoring, in particular to an AI vision recognition-based flame retardant particle quality online monitoring and sorting system. BACKGROUND

[0002] As a key additive of high polymer materials, the quality of flame retardant particles directly affects the fireproof performance of the final product. With the development of intelligent manufacturing technology, online monitoring systems based on AI vision recognition have gradually replaced traditional manual sampling inspection. Through high-resolution industrial cameras, the surface features (such as cracks, color differences, etc.) of the particles are captured, and deep learning algorithms are used to realize real-time sorting. However, for semi-transparent / transparent flame retardant particles (such as polycarbonate flame retardant masterbatch, optical grade flame retardant microspheres), conventional surface detection technology cannot penetrate the material interior. The current mainstream solution uses X-ray micro-focus tomography technology to identify defects based on density differences by reconstructing the internal structure of the particles in three dimensions.

[0003] Although the existing X-ray micro-focus tomography technology can analyze the internal structure, it still has obvious deficiencies in detecting refractive index abnormalities of transparent materials. Specifically, X-ray imaging relies on the physical principle of density difference of materials. The density difference between pure bubbles, transparent impurities, and the matrix material is extremely small, and the internal refractive index gradient change cannot be effectively characterized. Such defects lack significant contrast in X-ray images, resulting in high false negative rates for small transparent bubbles and cracks, and it is difficult to distinguish between internal real defects and surface optical artifacts, which seriously restricts the quality control requirements of high-precision transparent flame retardant materials. SUMMARY

[0004] The present application provides an AI vision recognition-based flame retardant particle quality online monitoring and sorting system, which separates surface noise by decomposing mixed polarization signals into azimuth and radial components, fuses feature band spectral data to construct an optical feature body, and finally generates a refractive index gradient field through a dual-channel network to analyze defect features, thereby solving the problems raised in the above background technology, i.e.:

[0005] It is difficult to accurately detect micron-level defects (such as pure bubbles and transparent cracks) inside semi-transparent flame retardant particles.

[0006] To achieve the above purpose, the flame retardant particle quality online monitoring and sorting system includes a polarized light field imaging module for acquiring and preprocessing the transmission image of the flame retardant particles to generate Stokes parameter images and multispectral transmittance distribution maps, and further includes a refractive index gradient field construction module and a dynamic sorting decision module.

[0007] The refractive index gradient field construction module is used for defect mechanism analysis on the refractive index gradient field to generate a defect three-dimensional heat map, and the defect three-dimensional heat map comprises:

[0008] A defect spatial coordinate layer for positioning a spatial physical position of the defect;

[0009] A type code layer for identifying a physical category of the defect;

[0010] A cause probability layer for associating the defect with a production process parameter;

[0011] The refractive index gradient field construction module comprises the following steps:

[0012] The Stokes parameter image is subjected to polarization feature decoupling processing to separate an azimuthal polarization component and a radial polarization component;

[0013] The multispectral transmittance distribution map is subjected to feature band extraction processing to screen feature band spectral data;

[0014] The radial polarization component and the feature band spectral data are fused to generate an optical feature body, and a double-channel network architecture is used to construct the refractive index gradient field;

[0015] The dynamic sorting decision module sorts out the flame retardant particles with internal defects based on the defect three-dimensional heat map.

[0016] In the above technical solution, the combination of multi-modal optical feature fusion and physical constraint modeling breaks through the dual limitations of traditional detection methods in information dimensionality and algorithm interpretability. If only relying on single polarization imaging, although the phase change of light waves can be analyzed, the interference of surface scattering on internal defect signals cannot be eliminated, making it difficult to distinguish between small bubbles and surface scratches. If only spectral analysis is used, although the material absorption characteristics can be reflected, the vector features of defect morphology are lost, and key information such as crack propagation direction cannot be obtained. The present solution removes surface noise by polarization feature decoupling, selects feature bands to enhance refractive index correlation, and constructs a multi-dimensional fusion optical feature body, thereby fundamentally solving the problem of extracting effective information from mixed signals. The double-channel network design further breaks through at the algorithm level: the optical feature channel retains the detail sensitivity of the measured data, the material characteristic channel embeds the inherent constraints of physical laws, and the two channels work together to avoid the excessive reliance on artificial experience of traditional threshold segmentation methods and overcome the failure risk of pure data-driven models for unknown defect types.

[0017] On this basis, the double-channel network architecture comprises an optical feature channel and a material characteristic channel, the optical feature channel processes polarization phase angle change data, and the material characteristic channel generates physical constraints in combination with feature band transmittance ratios and material intrinsic parameters.

[0018] In another technical solution, the defective machine analysis includes:

[0019] The refractive index gradient field construction module calculates the second derivative of each spatial point and locates the abnormal mutation area based on the vector matrix of the refractive index gradient field;

[0020] The refractive index gradient field construction module distinguishes radial defects and linearly arranged defects based on the gradient vector distribution pattern of the abnormal mutation area;

[0021] The refractive index gradient field construction module determines the physical origin position of the defect based on the direction of the gradient vector for back propagation calculation.

[0022] This technical solution overcomes the structural defects of traditional defect detection methods in feature characterization and cause tracing through the hierarchical design of dual-channel collaborative analysis and gradient field vector analysis. If only the optical feature channel is used, although the polarization phase change caused by the defect can be captured, process fluctuations may be misjudged as defects due to the lack of material physical constraints. If only the material property channel is used, although the physical law consistency can be ensured, the morphological characteristics of subtle defects may be lost, leading to the missed detection of micron-level cracks. The dual-channel verification mechanism of extracting the measured signal through the optical channel and injecting the physical priori through the material channel not only prevents the overfitting risk of pure data-driven, but also avoids the insufficient sensitivity of pure theoretical models. In the defect analysis stage, the abnormal area is located based on the basic defects of the second derivative calculation, the gradient vector pattern analysis realizes the intelligent classification of bubbles and cracks, and the back propagation tracing reflects the process causes of the defects, so that the system can not only find the existence of the defect, but also understand the nature of the defect.

[0023] Compared with the prior art, the beneficial effects of the present application are:

[0024] Through multi-dimensional analysis of the refractive index gradient field, the three-dimensional morphology reconstruction of the flame retardant particle quality defect, the physical cause inference, and the integrated diagnostic ability of the process correlation analysis are realized. Specifically, the construction of the gradient vector field not only reflects the geometric characteristics of the defect, but also reveals the potential correlation between the defect and the process conditions through the constraint of the material intrinsic parameters. The collaborative mechanism of the dual-channel network breaks through the dilemma between detection sensitivity and result reliability of traditional algorithms, so that the system has self-adaptive recognition ability for unknown defect types. This analysis paradigm deeply coupled with optical features, material properties and production processes provides complete technical support for flame retardant production from simple sorting to quality tracing, and realizes the transformation of quality control from "post-removal" to "process prevention". BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The overall flow structure of the present application is shown in the schematic diagram;

[0026] Figure 2 A flowchart of the polarized light field imaging module of the present application is shown in the figure.

[0027] Figure 3 A flowchart of the refractive index gradient field construction module of the present application is shown in the figure.

[0028] The meanings of the various labels in the figure are as follows:

[0029] 100, polarized light field imaging module; 200, refractive index gradient field construction module; 300, dynamic sorting decision module; 400, online self-optimization module. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0031] At present, it is difficult to accurately detect the micron-level defects (such as pure bubbles and transparent cracks) inside the semi-transparent flame retardant particles. The present application provides an AI vision-recognized flame retardant particle quality online monitoring and sorting system, as shown in Figure 1 The system includes a polarized light field imaging module 100, a refractive index gradient field construction module 200, a dynamic sorting decision module 300 and an online self-optimization module 400. Through the collaborative work of multiple modules, high-precision identification and automatic sorting of internal defects of semi-transparent flame retardant particles are realized.

[0032] The polarized light field imaging module 100 of the present application is located above the production line conveyor belt, and is mainly used to obtain high-quality optical characteristic data of semi-transparent flame retardant particles. The polarized light field imaging module 100 can penetrate the particle surface layer and capture internal optical information that cannot be obtained by traditional vision systems through a specially designed imaging system.

[0033] As shown in Figure 2 The core of the polarized light field imaging module 100 consists of three parts: an adjustable polarized light source system, a multi-spectral polarized camera array and a real-time image processing unit. The adjustable polarized light source system uses a ring-shaped LED array that can produce incident light of different polarization directions (0°, 45°, 90°, 135°), and can quickly switch the polarization state through precise control. The multi-spectral polarized camera array consists of four high-sensitivity industrial cameras, each equipped with a specific wavelength filter and a polarizer, and can synchronously capture transmission light images under different polarization states.

[0034] During operation, the flame retardant particles on the conveyor belt first pass through the polarized light source irradiation area. The system will emit incident light of four different polarization directions in turn, while the multispectral camera array synchronously collects the corresponding transmission images. This multi-angle, multi-spectral imaging method can maximize the extraction of the optical characteristic information inside the particles. The collected raw images are immediately transmitted to the real-time image processing unit for preprocessing.

[0035] The image processing unit performs three key processes on the raw data: first, background noise elimination to remove environmental light interference; then image registration to ensure that the images under different polarization states are strictly aligned; finally, synthesis of Stokes parameter images containing complete polarization information through a specific algorithm, and generation of multispectral transmittance distribution maps based on the transmission light intensity at different wavelengths. These processes ensure that the subsequent modules can obtain accurate and reliable input data.

[0036] After processing, the module outputs two key data sets: Stokes parameter images and multispectral transmittance distribution maps. The Stokes parameter images record the optical response characteristics of the particles under each polarization state, containing rich internal structure information; the multispectral transmittance distribution maps reflect the absorption characteristics of the particles to different wavelengths of light, calculated by the image processing unit by analyzing the transmission light intensity ratio at each wavelength.

[0037] It is particularly noteworthy that the module uses a unique ring-shaped light source design and a multi-camera synchronous acquisition scheme, ensuring that even at high-speed production line operation (particle movement speed up to 3 m / s), clear internal structure images can still be obtained. At the same time, the module's built-in self-calibration function can automatically compensate for imaging errors caused by changes in environmental temperature or equipment vibration, ensuring long-term stable operation.

[0038] However, these optical characteristic data alone cannot directly determine the internal defect situation of the particles. Although the Stokes parameter images record rich polarization information, the surface scattering effect contained therein can interfere with accurate judgment of internal defects; the multispectral transmittance distribution maps reflect the optical properties of the material, but lack precise description of the refractive index variation. In order to achieve accurate detection of internal defects, these optical characteristic data need to be further converted into physical quantity fields that can represent the internal structure of the material.

[0039] To this end, we introduce the refractive index gradient field construction module 200, which receives the Stokes parameter images and multispectral transmittance distribution maps processed by the polarized light field imaging module 100. The Stokes parameter images provide the optical response characteristics of the particles under different polarization states, and the multispectral transmittance distribution maps contain the light absorption characteristic information of the material, which together form the basis input for constructing the refractive index gradient field.

[0040] The refractive index gradient field construction module 200 is the core processing unit of the entire detection process, and its main function is to convert the optical characteristic data obtained by the polarized light field imaging module 100 into a physical quantity field that can accurately represent the internal structure of the material. Through innovative algorithm design, this module realizes accurate identification and positioning of internal defects in semi-transparent flame retardant particles.

[0041] As shown in Figure 3 The refractive index gradient field construction module 200 first receives two key input data from the polarized light field imaging module 100: the Stokes parameter image and the multispectral transmittance distribution map, and performs initial processing. In the initial processing stage, the module first decouples the polarization characteristics of the Stokes parameter image. Through vector decomposition algorithm, the mixed optical signal is accurately separated into azimuthal polarization component and radial polarization component. Among them, the azimuthal polarization component mainly reflects the scattering noise characteristics of the particle surface, while the radial polarization component reflects the refractive effect changes caused by internal defects. At the same time, the multispectral transmittance distribution map is processed by feature band extraction, and the feature band spectral data most related to the refractive index change of the material is selected.

[0042] Next, the refractive index gradient field construction module 200 deeply fuses the processed radial polarization component with the feature band spectral data to generate an optical feature body. This feature body accurately records the key optical characteristics of each position inside the particle, including polarization phase angle change data, feature wavelength transmittance ratio, and other core parameters. Through this fusion processing, the system effectively distinguishes between surface scattering effects and internal defect characteristics, ensuring that subsequent analysis is only directed at real internal defect signals, significantly improving detection accuracy. Among them, the polarization phase angle change data represents the birefringence effect caused by defects, and the feature wavelength transmittance ratio is directly related to the local refractive index change of the material. These two types of parameters together constitute the basic feature set for physical modeling.

[0043] For the polarization phase angle change data and feature wavelength transmittance ratio contained in the optical feature body, the module uses an innovative dual-channel network architecture to reconstruct the refractive index gradient field, in which: the optical feature channel (lower branch) is specifically designed to process polarization phase angle change data, and through deformable convolution layers, it adaptively extracts gradient features of particles of different sizes. The dynamically adjusted receptive field size of this channel directly depends on the spatial resolution recorded by the optical feature body, ensuring that the feature extraction accuracy strictly matches the characteristics of the input data. The material characteristic channel (upper branch) simultaneously uses the feature wavelength transmittance ratio and the pre-stored material intrinsic parameters (dielectric constant, magnetic permeability) to convert these physical quantities into physical constraints through a specially designed encoder. This processing makes the known correlation between the feature wavelength transmittance ratio and the local refractive index directly embedded in the network structure.

[0044] The dual-channel will give priority to retaining the measured gradient features extracted by the optical feature channel when fusing the features at the fusion layer, and the physical constraints generated by the material characteristic channel are used for verification and correction. This dual-verification mechanism ensures that the final output of the refractive index gradient field is both faithful to the measured data of the optical feature body and strictly in accordance with the physical laws of electromagnetic wave propagation.

[0045] The generated refractive index gradient field accurately characterizes the refractive index variation characteristics of each position inside the particle in the form of a three-dimensional vector matrix. Each spatial point corresponds to a three-dimensional vector, whose components represent the refractive index variation gradient of the point in the three orthogonal directions of the spatial rectangular coordinate system, and the vector modulus quantitatively represents the intensity of the local refractive index variation, and the vector direction reveals the energy diffusion trend of the defect region. This expression based on vector field not only breaks through the limitation of traditional grayscale image which can only reflect a single dimension, but also captures the three-dimensional expansion characteristics of defects through the spatial correlation of vectors, providing more rich physical information support for subsequent analysis.

[0046] For the constructed refractive index gradient field, the refractive index gradient field construction module 200 performs in-depth defect mechanism analysis. First, the second derivative of the gradient field is calculated based on the vector matrix of the refractive index gradient field, and the abnormal mutation region is accurately located by setting a dynamic threshold related to the elastic modulus of the material. Subsequently, based on the gradient vector distribution pattern of the abnormal mutation region, radial divergent defects and linearly arranged defects are distinguished: bubble defects usually present a radial divergent vector field distribution, while crack defects exhibit a linearly ordered vector arrangement. Finally, reverse propagation calculation is performed along the gradient vector direction to trace the origin position of the defect, and the possible causes are inferred in combination with the production process parameter database. Through these analyses, the system generates a three-dimensional thermal map of the defect containing three key information layers: the defect spatial coordinate layer accurately locates the abnormal region through three-dimensional coordinates, the type code layer identifies the defect category according to the gradient field feature pattern, and the cause probability layer evaluates the influence weight of each type of production parameter in combination with the process knowledge base. These structured data are directly transmitted to the dynamic sorting decision module 300 to provide comprehensive data support for automatic sorting. The entire processing flow is completed under the acceleration of an industrial-grade GPU, ensuring that the real-time requirements of the production line are met.

[0047] Since the three-dimensional thermal map of the defect output by the refractive index gradient field construction module 200 only contains the physical feature information of the defect, it cannot directly guide the sorting equipment to perform operations, and the present application designs a dynamic sorting decision module 300, which receives the three-dimensional thermal map of the defect from the upstream, including the defect spatial coordinate layer, the type code layer and the cause probability layer three types of structured information.

[0048] The dynamic sorting decision module 300 first performs coordinate system normalization processing on the defect space coordinate layer, converting the three-dimensional heat map coordinates into a two-dimensional mapping corresponding to the physical coordinates of the production line conveyor belt. In this process, the coordinate offset is dynamically adjusted in combination with the current conveyor belt speed to ensure that the positioning accuracy is not affected by the movement of the material. At the same time, the type code layer data is parsed by the classifier and converted into instruction codes recognizable by the sorting equipment, among which the bubble defects correspond to the gentle spraying mode, and the crack defects trigger the strong rejection instruction.

[0049] The core sorting process of the dynamic sorting decision module 300 adopts a multi-dimensional decision mechanism. The system first calculates the accurate sorting trigger time point based on the defect space coordinate layer data in combination with the real-time conveyor belt speed. At the same time, according to the defect category identified by the type code layer, the preset sorting intensity level is automatically matched. For defects marked as high process correlation in the cause probability layer, the system will preferentially allocate sorting resources to ensure that critical quality problems are handled in a timely manner.

[0050] The final generated sorting instruction set adopts a structured data format, including three core fields: the time control field records the millisecond-level precision trigger timestamp and action duration; the execution parameter field specifies the air valve spraying intensity level and target defect type code; the quality control field carries the detection confidence score and process association identifier. These instructions are transmitted in real time to the sorting execution mechanism through industrial Ethernet, and the transmission period is strictly controlled within 50ms to ensure timing accuracy.

[0051] A real-time monitoring mechanism for sorting effect is established when the system is running, continuously tracking key indicators such as instruction response delay, classification and rejection success rate, and mis-rejection rate. When the same type of defect is continuously not effectively rejected for 3 times, the system automatically starts the intensity upgrade protocol: first, increase the sorting intensity level of the same type of defect within the safety threshold, mark the related cases as optimization samples, and send a priority processing request to the online self-optimization module 400.

[0052] All sorting operations generate complete records with timestamps, which are synchronized in real time to the online self-optimization module 400 through a standardized interface. Each record contains four components: the original sorting instruction content, the execution device state parameters, the actual sorting effect verification result, and the system automatic adjustment log. These data and the corresponding original defect three-dimensional heat map, production line process parameter snapshot together constitute a closed-loop optimization data package, providing complete basis for continuous self-optimization of the system.

[0053] The online self-optimization module 400 receives the sorting instruction set and the corresponding defect three-dimensional heat map data transmitted by the dynamic sorting decision module 300 in real time as the intelligent optimization core of the system. The sorting instruction set completely records the execution parameters of each sorting operation, including the trigger time stamp accurate to milliseconds, the 1-8 adjustable air valve injection intensity level and the defect type code; and the received defect three-dimensional heat map contains the original detection data corresponding to this sorting, that is, the complete information of the spatial coordinate layer, the type code layer and the cause probability layer generated by the refractive index gradient field construction module 200. These data are transmitted in encrypted form through an industrial Ethernet, ensuring data integrity and timeliness.

[0054] Based on the received data, the online self-optimization module 400 performs intelligent optimization analysis: first, a sorting effect evaluation model is established to analyze the actual rejection effect of different types of defects under different sorting parameters, then the feature layer data of the defect three-dimensional heat map are combined to identify the optimization space of the existing sorting rules, and finally two sets of optimization schemes are generated: a double-channel network parameter update scheme is provided for the refractive index gradient field construction module 200 to optimize the defect detection accuracy; and a sorting rule adjustment suggestion is made for the dynamic sorting decision module 300 to improve the matching relationship between the intensity level and the defect type. All optimization instructions are fed back to the corresponding module through a standard interface to form a closed-loop optimization system that continuously improves itself.

[0055] In summary, the present application combines polarized optical detection and material physics modeling to generate a three-dimensional refractive index gradient field. This gradient field accurately quantifies the refractive index changes at each point inside the particle, enabling it to penetrate the material surface to detect pure bubbles and transparent cracks, and automatically distinguish defect types based on gradient distribution characteristics. This effectively solves the industry problem of accurately identifying transparent defects using traditional methods, and achieves comprehensive quality monitoring and sorting of flame retardant particles from the surface to the interior.

[0056] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited by the above examples, and the above examples and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. An online monitoring and sorting system for flame retardant particles based on AI visual recognition, comprising a polarization field imaging module (100), wherein the polarization field imaging module (100) is used to acquire transmitted light images of flame retardant particles and preprocess them to generate Stokes parameter images and multispectral transmittance distribution maps, characterized in that, It also includes a refractive index gradient field construction module (200) and a dynamic sorting decision module (300). The refractive index gradient field construction module (200) is used to perform defect mechanism analysis on the refractive index gradient field to generate a three-dimensional heat map of the defect, the three-dimensional heat map of the defect including: Defect space coordinate layer used to locate the physical position of defects in space; Type code layer used to identify the physical category of defects; A probability layer for linking defects to production process parameters; The refractive index gradient field construction module (200) includes the following steps for constructing the refractive index gradient field: Polarization feature decoupling processing is performed on the Stokes parameter image to separate the azimuth polarization component and the radial polarization component; The multispectral transmittance distribution map is processed to extract characteristic bands, and the spectral data of the characteristic bands are screened. The radial polarization component is fused with the characteristic band spectral data to generate an optical feature, and a refractive index gradient field is constructed using a dual-channel network architecture. The dynamic sorting decision module (300) sorts out flame retardant particles with internal defects based on the three-dimensional heat map of defects; The defect mechanism analysis includes: The refractive index gradient field construction module (200) calculates the second derivative of each spatial point and locates the abnormal abrupt change region based on the vector matrix of the refractive index gradient field; The refractive index gradient field construction module (200) distinguishes between radially divergent defects and linearly arranged defects based on the gradient vector distribution pattern of the anomalous abrupt change region; The refractive index gradient field construction module (200) performs backpropagation calculations based on the direction of the gradient vector to determine the physical origin location of the defect; The dynamic sorting decision module (300) receives the three-dimensional heat map of the defect and calculates the sorting trigger time point based on the defect spatial coordinate layer and the conveyor belt speed. The dynamic sorting decision module (300) determines the sorting intensity level based on the type code layer and generates a sorting instruction set containing time control field, execution parameter field and quality control field, which is transmitted to the sorting execution mechanism and the online self-optimization module (400).

2. The online monitoring and sorting system for flame retardant particles based on AI visual recognition as described in claim 1, characterized in that: The polarization field imaging module (100) includes an adjustable polarization light source system, which employs a ring-shaped LED array for switching polarization states.

3. The online monitoring and sorting system for flame retardant particles based on AI visual recognition according to claim 2, characterized in that: The polarization field imaging module (100) further includes a multispectral polarization camera array and a real-time image processing unit. The multispectral polarization camera array synchronously acquires transmitted light images under different polarization states. The real-time image processing unit performs background noise elimination and image registration on the transmitted light images, and synthesizes the Stokes parameter image and the multispectral transmittance distribution map.

4. The online monitoring and sorting system for flame retardant particles based on AI visual recognition according to claim 1, characterized in that: The optical feature records the optical properties of various locations inside the particle, including polarization phase angle variation data and transmittance ratios at specific wavelengths.

5. The online monitoring and sorting system for flame retardant particles based on AI visual recognition according to claim 1, characterized in that: The dual-channel network architecture includes an optical feature channel and a material property channel. The optical feature channel processes polarization phase angle variation data, and the material property channel combines the characteristic band transmittance ratio and the intrinsic material parameters to generate physical constraints.

6. The online monitoring and sorting system for flame retardant particles based on AI visual recognition according to claim 1, characterized in that: The online self-optimization module (400) receives the sorting instruction set and its corresponding three-dimensional heat map of defects, analyzes the correlation between sorting effect and defect characteristics, and generates a dual-channel network parameter update scheme and sorting rule adjustment suggestions.

7. The online monitoring and sorting system for flame retardant particles based on AI visual recognition according to claim 6, characterized in that: The online self-optimization module (400) transmits the network parameter update scheme to the refractive index gradient field construction module (200) and transmits the sorting rule adjustment suggestion to the dynamic sorting decision module (300).

Images