Intelligent recycling machine multi-modal garbage recognition method based on fuzzy classification

Abstract

The present application relates to object recognition technology field, specifically for the intelligent recycling machine multi-modal garbage identification method based on fuzzy classification, including the following steps: judging the garbage stable state through quality data, collecting garbage image, comparing image edge profile span and barycenter path, outputting profile offset feature information, combining power continuity and density to judge material type, adjusting feature channel priority, identifying garbage type and analyzing label conflict relationship, detecting abnormal sorting.In the present application, the stability of the garbage state is evaluated through the continuous quality change signal, the sampling opportunity is accurately determined, the deformation image accurate recognition is improved by combining the profile span difference and the barycenter offset trajectory, the material type is refined and classified by relying on the cross matching of the power continuity and the density data, and the classification abnormal garbage is identified and automatically isolated by judging the attribution offset of the classification label conflict, so that the recognition error caused by the unstable state, the shape change and the fuzzy material information is effectively avoided.

Description

Technical Field

[0001] This invention relates to the field of object recognition technology, and in particular to a multimodal waste recognition method for intelligent recycling machines based on fuzzy classification. Background Technology

[0002] The field of object recognition technology encompasses techniques for detecting, recognizing, and classifying targets contained in images or other forms of perceptual data. By extracting features and performing semantic understanding on input information such as images, videos, and sensor data, it determines the type, attributes, and location of target objects. This technology is significant in various application scenarios, including intelligent security, autonomous driving, medical image recognition, intelligent manufacturing, and intelligent recycling systems. In intelligent recycling systems, by recognizing the appearance of targets and combining it with non-visual attributes such as material, weight, and recyclability, it accurately classifies and determines the type of waste. This involves image processing models, feature fusion techniques, and multimodal data. The comprehensive application of analysis and auxiliary classification mechanisms includes a multimodal waste identification method for intelligent recycling machines based on fuzzy classification. This method integrates image recognition and non-visual feature analysis to comprehensively identify the complex and diverse attributes of household waste. It extracts the appearance and shape features of waste images through a lightweight Transformer structure, and then crops and optimizes the structure to reduce computational load. A fuzzy classification mechanism is introduced to fuzzify the material information and weight data of waste. A fuzzy rule base and membership judgment model are constructed, and the extracted features from the image are fused with the fuzzy classification output results in a multimodal manner. The waste category is determined through a feature association analysis unit.

[0003] Traditional waste identification technologies rely on single images or fixed data patterns for classification tasks when identifying objects. They ignore the changes in posture and state that may occur during waste disposal, and do not fully consider the impact of mass fluctuations and structural deformations on image accuracy. Furthermore, the fuzzy boundaries of non-visual information such as material and density of objects in the collected data are easily overlooked, leading to deviations in classification results. This can result in misclassification of items or failure to isolate abnormal items in a timely manner, reducing the accuracy and reliability of intelligent recycling systems. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides a multimodal waste identification method for intelligent recycling machines based on fuzzy classification. The technical solution is as follows: To achieve the above objectives, the present invention adopts the following technical solution: a multimodal waste identification method for intelligent recycling machines based on fuzzy classification, comprising the following steps: S1: Using a weighing sensor, analyze the continuous recorded mass change of the waste after it is disposed of, analyze the fluctuation of the mass data and determine the stable state of the items, send an image sampling command to the image acquisition device, and output a waste image dataset. S2: Using the aforementioned garbage image dataset, compare the differences between the maximum and minimum horizontal spans of the edge contour points of continuous images in the horizontal direction, analyze the connection paths of the centroid coordinates of multiple locations in the grayscale region of the image under horizontal projection, evaluate the degree of deformation of the image structure, and output contour offset feature information. S3: Based on the contour offset feature information, calculate the continuity of the conductive probe's current during contact with the target, combine the laser volume profiler and weight information to calculate the density information, output the fuzzy label code that matches the waste material type label, and generate the attribution level code; S4: Call the attribution level code to determine the clarity and coherence of the image edge structure, analyze the update response times of the corresponding channel in the encoder for the activated region, adjust the update priority of the target feature channel, identify the type label of the garbage, and generate the garbage type identification result.

[0005] As a further aspect of the present invention, the garbage image dataset includes image acquisition start time, reflection response difference spectrum, and image stability judgment label. The contour offset feature information specifically includes lateral span comparison parameters, centroid connection path change data, and image confidence score. The attribution level coding includes power continuity classification code, volume density combination level, and material label fuzzy coding. The garbage type identification result specifically refers to edge structure clarity identifier, coding channel response frequency, and type label index.

[0006] As a further aspect of the present invention, the steps for obtaining the garbage image dataset are specifically as follows: S101: Call the weighing sensor to acquire the mass measurement signal, analyze the mass change trend between multiple sampling times, evaluate the stable state of the waste in real time, and establish the stable state judgment result; S102: Based on the stable state determination result, identify the static state of the garbage and send an image sampling command to the image acquisition device to acquire multiple frames of garbage images, call the pixel value data of the specified image area, calculate the reflection response change of the same area under multiple image frames, and obtain the reflection change coefficient. S103: Based on the reflection change coefficient, evaluate the stability of image sampling, detect sampling anomalies and output a resampling command to obtain a garbage image dataset.

[0007] As a further aspect of the present invention, the specific formula for calculating the reflection response change of the same region under multiple image frames is as follows: ; Calculate the reflection variation coefficient; in, The reflection variation coefficient, For pixel index number, Number the image frames. For the first The first frame of the image grayscale value of each pixel. For the first The first frame of the image grayscale value of each pixel. For the first The normalized value of the horizontal pixel gradient of each pixel. For the first The normalized grayscale gradient value of each pixel. The total number of pixels selected.

[0008] As a further aspect of the present invention, the step of obtaining the contour offset feature information specifically includes: S201: Obtain the garbage image dataset, analyze the coordinates of edge contour points in each frame image, compare the difference between the maximum and minimum horizontal span on the horizontal axis, and generate a span difference index. S202: Based on the span difference index, select the centroid coordinates of multiple locations in the grayscale region of the image, connect each centroid coordinate point to construct a horizontal projection path, analyze the direction of change and the degree of offset of the path, and output the centroid trajectory change parameters. S203: Based on the changes in the center of gravity trajectory, analyze the deformation trend, combine the span difference index and the range of changes in the center of gravity trajectory, analyze the degree of deformation of the image structure, calculate the confidence level of the image, filter the image, and obtain contour offset feature information.

[0009] As a further aspect of the present invention, the step of obtaining the attribution level code specifically includes: S301: Obtain the contour offset feature information, collect the current response signal of the conductive probe during the contact process with the target, analyze the continuous performance of current on and off, and output the power continuity data. S302: Based on the power continuity data, obtain the three-dimensional structural data measured by the laser volume profiler and the mass data of the weighing sensor, calculate the volume and density of the object, and obtain the cross-state parameters by combining the power continuity. S303: Based on the cross-state parameters, filter the waste material type labels that match the current parameter combination, output the fuzzy label code that matches the waste material type label, and output the attribution level code.

[0010] As a further aspect of the present invention, the specific formula for filtering waste material type tags that match the current parameter combination is as follows: ; Calculate the matching difference index; in, This is the normalized value for the continuity of power supply of the current target waste. The density normalized value of the current target waste. For the first Normalized mean of electrical continuity of waste of known material type For the first The normalized mean density of waste of known material type. For the first The variance of normalized samples of known material electrical continuity. For the first The variance of a normalized sample of known material density. For the first Normalized weights for classification confidence scores of known materials. Used to retrieve the index value corresponding to the minimum value of the objective function. For each type of known material waste in the sample library, use a type number index. This represents the total number of categories of known material waste in the sample library. The result is the known material label number that has the smallest difference from the current target waste.

[0011] As a further aspect of the present invention, the step of obtaining the garbage type identification result specifically includes: S401: Call the attribution level code to calculate the gray-level gradient of each edge structure distribution area in the image, determine the sharpness level and edge coherence of each area, and obtain the structural sharpness index. S402: Based on the structural clarity index, analyze the response update count of the channel corresponding to each active region in the encoder, adjust the priority of the target region feature data in the channel and the participation frequency of adjacent regions, and establish feature priority parameters; S403: Based on the feature priority parameter, identify the type of target waste, match the type label, and generate waste type identification results.

[0012] As a further aspect of the present invention, the method further includes: S5: Call the waste type identification result and the classification level code, determine the conflict relationship of the classification labels in the material classification direction, analyze the degree of classification offset between classification labels and calculate the degree of conflict in the classification direction, identify abnormal waste, output isolation sorting signal, and obtain abnormal sorting control instructions. The abnormal sorting control instructions include classification direction offset, belonging label conflict level, and sorting path isolation signal.

[0013] As a further aspect of the present invention, the step of obtaining the abnormal sorting control instruction specifically includes: S501: Call the waste type identification result and the classification level code, determine the matching relationship between the two classification labels in the material classification direction, analyze the classification offset between the labels, calculate the conflict value of the classification direction, and generate classification conflict parameters. S502: Based on the aforementioned classification conflict parameters, identify and filter abnormally classified waste, output an isolation sorting signal and transmit it to the waste sorting equipment, and obtain abnormal sorting control instructions.

[0014] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following: By assessing the stability of waste status through continuous quality change signals, the sampling timing can be accurately determined. By combining the difference in contour span with the center of gravity offset trajectory, the accuracy of deformation image recognition can be improved. By relying on the continuity of power supply and cross-matching of density data, the material type can be refined and classified. By judging the attribution offset of conflicting classification labels, abnormal waste can be identified and automatically isolated, effectively avoiding recognition errors caused by unstable disposal status, shape changes and ambiguity of material information. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the workflow of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Please see Figure 1 This invention provides a technical solution: a multimodal waste identification method for intelligent recycling machines based on fuzzy classification, comprising the following steps: S1: Using a weighing sensor, analyze the continuous recorded mass change of the waste after it is disposed of, analyze the fluctuation of the mass data and determine the stable state of the items, send an image sampling command to the image acquisition device, and output a waste image dataset. S2: Using a garbage image dataset, compare the differences between the maximum and minimum horizontal spans of continuous image edge contour points in the horizontal direction, analyze the connection paths of the centroid coordinates of multiple locations in the gray area of ​​the image under horizontal projection, evaluate the degree of image structure deformation, and output contour offset feature information. S3: Based on the contour offset feature information, calculate the continuity of the conductive probe's current during contact with the target, combine the laser volume profiler and weight information to calculate the density information, output the fuzzy label code that matches the waste material type label, and generate the attribution level code; S4: Call the attribution level coding to determine the clarity and coherence of the image edge structure, analyze the update response times of the corresponding channels in the encoder for the activated region, adjust the update priority of the target feature channels, identify the type label of the garbage, and generate the garbage type identification result; S5: Call the waste type identification results and classification level code, determine the conflict relationship of classification labels in the material classification direction, analyze the degree of classification offset between classification labels and calculate the degree of conflict in classification direction, identify abnormal waste, output isolation sorting signal, and obtain abnormal sorting control instructions.

[0023] The waste image dataset includes image acquisition start time, reflection response difference map, image stability judgment label, contour offset feature information specifically including lateral span comparison parameters, centroid connection path change data, image confidence score, attribution level coding including power continuity classification code, volume density combination level, material label fuzzy coding, waste type identification results specifically including edge structure clarity identifier, coding channel response frequency, type label index, and abnormal sorting control instructions including classification direction offset, attribution label conflict level, and sorting path isolation signal.

[0024] The specific steps for obtaining the garbage image dataset are as follows: S101: Call the weighing sensor to acquire the mass measurement signal, analyze the mass change trend between multiple sampling times, evaluate the stable state of the waste in real time, and establish the stable state judgment result; The system uses a weighing sensor to acquire continuous mass measurement data of waste falling onto the platform at 0.5-second intervals. By constructing a horizontal time series, the system calculates the fluctuation in the difference between adjacent data points. In the initial stage of detection, there are usually sharp fluctuations in the readings due to inertial impact. At this time, three stable trend segments are extracted using a moving average method. The standard deviation of the sampling points in each segment is calculated to determine whether a stable period has been entered. When the standard deviation of three consecutive segments is lower than the set range, it is considered to be in a stable state. For example, if the lower limit mass is set to 100 grams, and the continuously measured sequence is 523g, 519g, 517g, 518g, 517g, the fluctuation rate is calculated to determine whether it is trending towards stability. During this process, the system automatically records the difference and fluctuation amplitude of each sampling point and displays its trend changes in real time on the graphical interface. The table below shows the difference change pattern under a stable trend. Table 1. Trend of Quality Sampling Fluctuations

[0025] As shown in Table 1, the fluctuation trend from t3 to t5 is stable, the system recording status switches to "images that can be acquired", and outputs the stable state determination result.

[0026] S102: Based on the steady state determination result, identify the static state of the garbage and send an image sampling command to the image acquisition device to acquire multiple frames of garbage images, call the pixel value data of the specified image area, calculate the reflection response change of the same area under multiple image frames, and obtain the reflection change coefficient. The specific formula for calculating the reflection response changes of the same region across multiple image frames is as follows: ; Calculate the reflection variation coefficient; in, The reflection variation coefficient, For pixel index number, Number the image frames. For the first The first frame of the image grayscale value of each pixel. For the first The first frame of the image grayscale value of each pixel. For the first The normalized value of the horizontal pixel gradient of each pixel. For the first The normalized grayscale gradient value of each pixel. The total number of pixels selected.

[0027] formula: ; Detailed explanation of the formula and its calculation derivation: The formula is used to calculate the change in reflection response of pixels in the same region in two adjacent frames of images. The result is used to measure the optical stability during the image sampling process and to determine whether the image has reached a stable state that can be used for subsequent recognition processing.

[0028] Parameter meanings and settings: This represents the grayscale value of the i-th pixel in the t-th frame of the image. It is acquired by the image acquisition device and read through the pixel matrix. Under the background of varying brightness in the appearance of the garbage, the value ranges from 20 to 240. , , ; This represents the grayscale value at the same pixel position in the (t+1)th frame of the image. The device captures the image with a time delay. , , ; The horizontal pixel gradient normalization value of the i-th pixel is obtained by normalizing the absolute value of the slope of the horizontal gray-level change in the image, with reference to the maximum gradient setting of 255. , , ; The grayscale gradient normalization value of the i-th pixel is obtained by normalizing the average gradient of its 8 neighboring pixels. , , ; The number of pixels involved in the calculation is set to 3; Substitute the parameters into the formula to calculate: ; ; ; ; The result of 0.0308 indicates the average relative change in the reflection response between adjacent image frames within the sampling area. The closer the value is to zero, the more stable the optical response of the image sampling is. This value, as the reflection change coefficient, will be used as the basis for subsequent determination of image stability and screening of image sampling control.

[0029] S103: Based on the reflection change coefficient, evaluate the stability of image sampling, detect sampling anomalies and output resampling instructions to obtain a garbage image dataset; Based on the reflection change coefficient, the system performs an overall evaluation of the grayscale response changes of the three frames. Taking the current frame group as the evaluation unit, the system calculates the mean normalization of the RGB channel mean changes in each data group to assess the overall imaging jitter interference. If the average fluctuation of any of the three channels exceeds the specified standard, the image frame group is marked as an unstable sampling group. For example, in image frame group B, the average fluctuation of the G channel is 4.2%, the R channel is 3.6%, and the B channel is 2.9%, with a combined average fluctuation exceeding 3.5%, which does not meet the set standard. In this case, the system calls a resampling command to control the channel to re-enter the exposure state and acquire image frames. If all channels are within the standard after evaluation, the system uses the image of this frame group as one of the input sources. Then, it combines the timestamp to organize the file structure and number the frame group, generating a multimodal image input record and finally obtaining the garbage image dataset.

[0030] The specific steps for obtaining contour offset feature information are as follows: S201: Obtain a dataset of garbage images, analyze the coordinates of edge contour points in each frame of the image, compare the difference between the maximum and minimum horizontal span on the horizontal axis, and generate a span difference index. After acquiring the garbage image dataset, the system first performs edge extraction on each frame of the image. Using the Sobel edge extraction template, it calculates the gradient of each pixel in the image, locates areas with significant gray-level jumps in the edge map, and uses these as candidate contour point regions. By scanning line by line, it extracts the coordinates of the leftmost and rightmost edge points in each row, calculates the horizontal span of each row, and then searches for the maximum and minimum horizontal spans in the entire frame to construct a set of horizontal spans. The difference in this set is used to calculate the discrete fluctuation range of the overall contour. For example, in a certain image frame, the maximum horizontal span is 156 pixels, and the minimum span is 132 pixels, then the span difference is 24 pixels. This value will be included in a standardized evaluation table for normalization. The table below lists the horizontal span data and their differences collected from different image samples: Table 2 Statistical Table of Horizontal Span Differences

[0031] As shown in Table 2, the difference range in this image sequence is between 20 and 26 pixels, and a span difference index is generated according to the set grading standard.

[0032] S202: Based on the span difference index, select the centroid coordinates of multiple locations in the grayscale region of the image, connect each centroid coordinate point to construct a horizontal projection path, analyze the direction of change and the degree of offset of the path, and output the centroid trajectory change parameters. Based on the span difference index, the system extracts five equally spaced location regions from the image grayscale as centroid analysis points. Within each region, the weighted average coordinates of the pixel grayscale values ​​are calculated to represent the grayscale centroid of that region. These five centroid coordinates are then sequentially connected to construct a two-dimensional horizontal projection path. The relative coordinate differences between adjacent centroids and the total offset of the overall path are recorded along this path. The angle distribution between the line connecting the first and last points and each path segment is calculated to obtain the path's trend. If the path is skewed downwards, the system records the offset direction as a negative trend and constructs a centroid change description vector using the mean angle and the total coordinate offset. For example, if the mean offset angle of an image path is 8.6 degrees, the horizontal centroid offset is 19 pixels, and the vertical offset is 12 pixels, this vector will be stored as a standard feature vector for subsequent use by the deformation discrimination unit, and the centroid trajectory change parameters will be output.

[0033] S203: Based on the change parameters of the center of gravity trajectory, analyze the deformation trend, combine the span difference index and the change range of the center of gravity trajectory, analyze the degree of deformation of the image structure, calculate the confidence of the image and filter the image to obtain contour offset feature information. Based on the center-of-gravity trajectory change parameters, the system analyzes the overall offset shape of the five-point center-of-gravity path, combines the path offset, average angle, and span difference index to form a structural deformation reference vector, and calls the normalized threshold region to classify the deformation. When the average path angle is greater than 10 degrees or the total lateral offset exceeds 15 pixels, the system marks the image as a structurally abnormal input. If the corresponding span difference index also exceeds the baseline difference range, the system marks the image as "insufficient reconstruction confidence". Taking image IMG03 as an example, its span difference is 26 pixels, the center-of-gravity path offset is 19 pixels laterally, and the average angle is 11.3 degrees. According to the rules, the deformation degree of this image is judged to be high, and a weight value of 0.42 is assigned to it in the confidence matrix. The system establishes a confidence distribution sequence based on the score records of all scoring frames, and filters image frames with a confidence of less than 0.5 as invalid frames. At the same time, the deformation parameters and structural contours of the valid frames are jointly marked to finally obtain the contour offset feature information.

[0034] The specific steps for obtaining the attribution level code are as follows: S301: Acquire contour offset feature information, collect current response signals of conductive probe during contact with target, analyze the continuous performance of current on / off, and output energization continuity data. After acquiring the contour offset feature information, the system synchronously activates the conductivity detection device, placing the conductive probe in contact with the surface of the object corresponding to the image at a constant pressure. The system records the current response signal during the continuous time period from the start of probe contact to disconnection, and collects the current value change curve in the signal time series. Mean filtering is performed at 0.01-second intervals for each time period to determine whether the current is consistently and stably greater than the reference conduction current value of 5mA. The system accumulates the effective time of the continuous current to obtain the effective conduction duration within the complete contact cycle. Simultaneously, it analyzes the number of current intermittent fluctuations and the fluctuation intervals, constructs a conduction stability vector, and establishes a classification conductivity level, setting it to three levels: complete conduction, weak conduction, and intermittent conduction. The continuity data of the current is output according to the level. To further compare the conductivity of different samples, the system statistically analyzes the detection time, mean current, and number of intermittent currents for multiple waste samples, as shown in the table below: Table 3. Results of Conductivity Test on Waste Samples

[0035] Referring to Table 3, the system is classified as follows: A01 is fully conductive, A02 is weakly conductive, and A03 is intermittently conductive, which serves as the input basis for subsequent state judgment.

[0036] S302: Based on the continuity of power supply data, acquire the three-dimensional structural data measured by the laser volume profiler and the mass data of the weighing sensor, calculate the volume and density of the object, and obtain the cross-state parameters by combining the continuity of power supply. Based on the aforementioned continuity data, the system synchronously calls a laser volume profiler to acquire the object's 3D point cloud data. After reconstruction, its length, width, and height boundary parameters are extracted, and the estimated volume is obtained using the circumscribed cube volume calculation method. Simultaneously, the system reads the mass data recorded by the weighing sensor within the corresponding time period, performs a ratio calculation between the volume value and the mass data, and calculates the unit volume density value. For example, sample A02 is 12.4cm long, 7.8cm wide, and 3.1cm high, with a volume of 300.1cm³ and a mass of 152.3g, corresponding to a density of 0.507g / cm³. The system performs a cross-combination mapping of conductivity level and density range to construct cross-state parameters in the 3D attribute space. If the density is in the medium density range (0.4–0.8g / cm³) and the conductivity is weak, the system defines its cross-state as the composite material neutral label input and outputs this result as the cross-state parameter.

[0037] S303: Based on the cross-state parameters, filter the garbage material type labels that match the current parameter combination, output the fuzzy label code that matches the garbage material type label, and output the attribution level code; The specific formula for filtering out garbage material type tags that match the current parameter combination is: ; Calculate the matching difference index; in, This is the normalized value for the continuity of power supply of the current target waste. The density normalized value of the current target waste. For the first Normalized mean of electrical continuity of waste of known material type For the first The normalized mean density of waste of known material type. For the first The variance of normalized samples of known material electrical continuity. For the first The variance of a normalized sample of known material density. For the first Normalized weights for classification confidence scores of known materials. Used to retrieve the index value corresponding to the minimum value of the objective function. For each type of known material waste in the sample library, use a type number index. This represents the total number of categories of known material waste in the sample library. The result is the known material label number that has the smallest difference from the current target waste.

[0038] formula: ; Detailed explanation of the formula and its calculation derivation: The formula is used to calculate the attribute matching difference value between the target waste and samples of different known materials. By comparing the normalized offset of electrical continuity and density, and combining the attribute fluctuation range and confidence weight of the material samples, the label number corresponding to the minimum difference value is obtained, and the fuzzy label code and attribution level code matched by the number are output. Parameter meanings and settings: The normalized value for the continuity of electrification of the target waste is set to 0.68, which reflects the stability of the target's conductivity response per unit time. The density normalization value for the target waste is set to 0.75; For the first The average value of the continuity of current carrying capacity of the material, with a selected metal sample. ; This represents the average density of this type of material. ; Let the variance of the continuity of current flow in metallic classes be denoted as . The mean square error is derived from the power-on data of similar samples. Let the density variance of metallic compounds be denoted as... Calculate the density difference of samples from the same category; The confidence weight for identifying metallic materials is derived from the stability evaluation of this type of sample identification and is set to 0.9. The weight is normalized according to the confidence interval, with a full score of 1. A value of 0.9 is assigned to those with a matching frequency of more than 9 times in the last 10 identifications. Substitute the parameters into the objective function within the parentheses in the formula to calculate the matching difference index value for each type of known material, and denote the value of the first... The matching difference index value of the material type is : Similarly, to calculate the label for the second category of plastics, let... , , , , The calculation is as follows: (Plastics, ); The results show , Matching difference index in metallic materials ( To obtain the minimum value, based on The index corresponding to the minimum value of the objective function is obtained. This means outputting the metal material label number; indicating that the target waste is most similar to the metal sample in terms of electrical conductivity and density attributes. Finally, the fuzzy label code corresponding to this type of material is identified and matched, completing the grade coding and labeling of the material classification.

[0039] The specific steps for obtaining the garbage type identification results are as follows: S401: Call the attribution level code, calculate the gray-level gradient for each edge structure distribution area in the image, determine the sharpness level and edge coherence of each area, and obtain the structural sharpness index. The attribution level coding is invoked to calculate the gray-level gradient for each edge structure distribution region in the image. First, the gray-level differences between adjacent pixels in the horizontal and vertical directions are obtained from the image pixel matrix. Then, based on the contour boundary coordinates of each region, the gradient magnitude of adjacent points is accumulated, and the average gray-level gradient magnitude within a unit length of edge is calculated. Regions with gray-level gradients greater than the edge sharpness benchmark are selected, and the average density level of gray-level changes within these regions is recorded. Further, the connectivity matrix of each edge region is invoked, and the edge coherence coefficient is calculated based on the pixel consistency and structural closure between adjacent boundary nodes. This coefficient is compared with an internally preset edge coherence structure threshold to determine the edge continuity level of each region. The average gray-level gradient and edge coherence coefficient are combined to calculate the structural sharpness index for each edge structure region. For ease of explanation, the following are the sharpness evaluation results of different region samples in multiple images: Table 4. Image Edge Region Structural Sharpness Index

[0040] As shown in Table 4, different image regions can obtain different sharpness indices based on grayscale changes and edge connectivity, which are used as the basis for subsequent ranking of the importance of regional features.

[0041] S402: Based on the structural clarity index, analyze the response update count of each channel corresponding to each active region in the encoder, adjust the priority of the target region feature data in the channel and the participation frequency of adjacent regions, and establish feature priority parameters; Based on the structural clarity index, the response records corresponding to the active regions in the encoder within the feature extraction channels are analyzed sequentially. First, the response update count of each feature channel during the processing of the current image is obtained. The encoder is an image feature extraction encoder deployed in the recycling machine controller and built based on the existing lightweight Transformer structure. It consists of an image block embedding layer, a position coding layer, several cascaded multi-head self-attention layers and feedforward network layers, and a layer normalization module. The feature extraction channel refers to the feature mapping channel output by each attention head in the multi-head self-attention layer. The active region refers to the image block region that is significantly activated under the action of self-attention weights after the input image is block-embedded. The response update count is the number of times the corresponding channel is activated and the feature value is updated in the forward computation of consecutive frames. The channel activity parameter is obtained by statistically analyzing the average number of activations of the corresponding region in consecutive frames of images for each channel. This is then compared with the sharpness index of the corresponding image region. When a channel region has a sharpness index value higher than the average standard and the number of activations reaches the channel activity benchmark value, the channel region is marked as a high-priority channel. Subsequently, the adjacency frequency of adjacent regions in the current feature channel set is statistically analyzed. The statistical value is calculated by the number of connecting edges and the percentage of shared pixels between channels. Finally, all feature channels are divided into three priority levels: high, medium, and low. A two-factor matrix of channel priority and adjacency frequency is established to generate the feature priority parameters for each feature region.

[0042] S403: Based on the feature priority parameter, identify the type of target waste, match the type label, and generate waste type identification results; Based on the feature priority parameters, the target regions in each feature channel are classified and judged one by one. First, the response region in the high-priority channel is called, and the region features are compared with the pattern contour. The structural similarity coefficient is used to compare the current region contour with the standard contour samples in the waste identification database. The structural similarity coefficient is the joint comparison result of gray distribution and boundary shape. The evaluation interval of matching confidence level is set. If the channel region structural similarity coefficient is greater than the matching benchmark, it is judged as a classable target. The region label information and the classification level code are cross-compared to determine the classification orientation of the current waste type in the existing classification system. If the classification is consistent, the corresponding waste type label is output. If the classification offset is greater than the classification adjustment standard, it is reclassified into the category with low label confidence. All image targets that have been identified are labeled and output. Finally, all valid target recognition results are integrated to generate waste type recognition results.

[0043] The specific steps for obtaining abnormal sorting control instructions are as follows: S501: Call the waste type identification result and the classification level code, determine the matching relationship between the two classification labels in the material classification direction, analyze the classification offset between the labels, calculate the conflict value of the classification direction, and generate classification conflict parameters. The system retrieves the garbage type identification results and classification level codes, selects the main label and its associated material attributes from the current identification results, parses the material orientation information carried by the classification level code, constructs a classification direction coordinate set by comparing the classification reference vector of the main label with the direction attribute sequence in the classification level code, analyzes the number of main axis offsets of the two types of labels in this set, extracts the intersection segments between various types of labels using a one-dimensional projection method, and calculates the percentage segment width of the intersection area within the total projection width to obtain the offset ratio value. At the same time, it obtains the weight score of all material mapping combinations between the main label and the classification code label, and constructs the classification conflict value based on the product of the offset ratio and the combination score. It summarizes the coordinate point distribution characteristics of this value under the set classification logical coordinate system and constructs the classification conflict parameters. The table below shows the conflict value statistics of various label combinations in the classification conflict parameters.

[0044] Table 5. Parameter Table for Conflicts in Attribution Direction Label Combinations

[0045] The closer the value of the attribution conflict parameter is to the error zone in the attribution logic model, the more it indicates that the attribution matching state between labels deviates from the principle combination, which is prone to triggering misclassification.

[0046] S502: Based on the classification conflict parameters, identify and filter abnormally classified waste, output isolation sorting signals and transmit them to the waste sorting equipment, and obtain abnormal sorting control instructions; Based on the attribution conflict parameters, the attribution matching score of each sample in the current batch is extracted from the waste type identification results. A screening threshold is set according to the degree of offset of the attribution conflict parameters in the coordinate set, and all sample numbers with offsets exceeding the allowable matching boundary are screened out. The physical feature index code of the screened sample is called, and an instruction carrier is generated through index mapping. A binary isolation matrix is ​​constructed in the control platform, and the control path switching is triggered by the matrix state. The isolation sorting execution component is activated, and the synchronous distribution instructions of the steering device and the air pressure baffle are linked to exclude abnormal classification targets from the standard delivery path. At the same time, the screened number and its mapping address are pushed into the pending instruction pool, and the sorting main control module issues an independent isolation control command to the waste sorting equipment through the CAN communication protocol to obtain the abnormal sorting control instruction.

[0047] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0048] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0049] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0050] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0051] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0052] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0053] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0054] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0055] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0056] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0057] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multimodal waste identification method for intelligent recycling machines based on fuzzy classification, characterized in that, The method includes: S1: Using a weighing sensor, analyze the continuous recorded mass change of the waste after it is disposed of, analyze the fluctuation of the mass data and determine the stable state of the items, send an image sampling command to the image acquisition device, and output a waste image dataset. S2: Using the aforementioned garbage image dataset, compare the differences between the maximum and minimum horizontal spans of the edge contour points of continuous images in the horizontal direction, analyze the connection paths of the centroid coordinates of multiple locations in the grayscale region of the image under horizontal projection, evaluate the degree of deformation of the image structure, and output contour offset feature information. S3: Based on the contour offset feature information, calculate the continuity of the conductive probe's current during contact with the target, combine the laser volume profiler and weight information to calculate the density information, output the fuzzy label code that matches the waste material type label, and generate the attribution level code; S4: Call the attribution level code to determine the clarity and coherence of the image edge structure, analyze the update response times of the corresponding channel in the encoder for the activated region, adjust the update priority of the target feature channel, identify the type label of the garbage, and generate the garbage type identification result.

2. The multimodal waste identification method for intelligent recycling machines based on fuzzy classification according to claim 1, characterized in that, The garbage image dataset includes image acquisition start time, reflection response difference map, and image stability judgment label. The contour offset feature information specifically includes lateral span comparison parameters, centroid connection path change data, and image confidence score. The attribution level coding includes power continuity classification code, volume density combination level, and material label fuzzy coding. The garbage type identification result specifically refers to edge structure clarity identifier, coding channel response frequency, and type label index.

3. The multimodal waste identification method for intelligent recycling machines based on fuzzy classification according to claim 1, characterized in that, The specific steps for obtaining the garbage image dataset are as follows: S101: Call the weighing sensor to acquire the mass measurement signal, analyze the mass change trend between multiple sampling times, evaluate the stable state of the waste in real time, and establish the stable state judgment result; S102: Based on the stable state determination result, identify the static state of the garbage and send an image sampling command to the image acquisition device to acquire multiple frames of garbage images, call the pixel value data of the specified image area, calculate the reflection response change of the same area under multiple image frames, and obtain the reflection change coefficient. S103: Based on the reflection change coefficient, evaluate the stability of image sampling, detect sampling anomalies and output a resampling command to obtain a garbage image dataset.

4. The multimodal waste identification method for intelligent recycling machines based on fuzzy classification according to claim 3, characterized in that, The specific formula for calculating the reflection response changes of the same region across multiple image frames is as follows: ； Calculate the reflection variation coefficient; in, The reflection variation coefficient, For pixel index number, Number the image frames. For the first The first frame of the image grayscale value of each pixel. For the first The first frame of the image grayscale value of each pixel. For the first The normalized value of the horizontal pixel gradient of each pixel. For the first The normalized grayscale gradient value of each pixel. The total number of pixels selected.

5. The multimodal waste identification method for intelligent recycling machines based on fuzzy classification according to claim 3, characterized in that, The specific steps for obtaining the contour offset feature information are as follows: S201: Obtain the garbage image dataset, analyze the coordinates of edge contour points in each frame image, compare the difference between the maximum and minimum horizontal span on the horizontal axis, and generate a span difference index. S202: Based on the span difference index, select the centroid coordinates of multiple locations in the grayscale region of the image, connect each centroid coordinate point to construct a horizontal projection path, analyze the direction of change and the degree of offset of the path, and output the centroid trajectory change parameters. S203: Based on the changes in the center of gravity trajectory, analyze the deformation trend, combine the span difference index and the range of changes in the center of gravity trajectory, analyze the degree of deformation of the image structure, calculate the confidence level of the image, filter the image, and obtain contour offset feature information.

6. The multimodal waste identification method for intelligent recycling machines based on fuzzy classification according to claim 5, characterized in that, The specific steps for obtaining the attribution level code are as follows: S301: Obtain the contour offset feature information, collect the current response signal of the conductive probe during the contact process with the target, analyze the continuous performance of current on and off, and output the power continuity data. S302: Based on the power continuity data, obtain the three-dimensional structural data measured by the laser volume profiler and the mass data of the weighing sensor, calculate the volume and density of the object, and obtain the cross-state parameters by combining the power continuity. S303: Based on the cross-state parameters, filter the waste material type labels that match the current parameter combination, output the fuzzy label code that matches the waste material type label, and output the attribution level code.

7. The multimodal waste identification method for intelligent recycling machines based on fuzzy classification according to claim 6, characterized in that, The specific formula for filtering waste material type tags that match the current parameter combination is as follows: ； Calculate the matching difference index; in, This is the normalized value for the continuity of power supply of the current target waste. The density normalized value of the current target waste. For the first Normalized mean of electrical continuity of waste of known material type For the first The normalized mean density of waste of known material type. For the first The variance of normalized samples of known material electrical continuity. For the first The variance of a normalized sample of known material density. For the first Normalized weights for classification confidence scores of known materials. Used to retrieve the index value corresponding to the minimum value of the objective function. For each type of known material waste in the sample library, use a type number index. This represents the total number of categories of known material waste in the sample library. The result is the known material label number that has the smallest difference from the current target waste.

8. The multimodal waste identification method for intelligent recycling machines based on fuzzy classification according to claim 6, characterized in that, The specific steps for obtaining the waste type identification result are as follows: S401: Call the attribution level code to calculate the gray-level gradient of each edge structure distribution area in the image, determine the sharpness level and edge coherence of each area, and obtain the structural sharpness index. S402: Based on the structural clarity index, analyze the response update count of the channel corresponding to each active region in the encoder, adjust the priority of the target region feature data in the channel and the participation frequency of adjacent regions, and establish feature priority parameters; S403: Based on the feature priority parameter, identify the type of target waste, match the type label, and generate waste type identification results.

9. The multimodal waste identification method for intelligent recycling machines based on fuzzy classification according to claim 1, characterized in that, The method further includes: S5: Call the waste type identification result and the classification level code, determine the conflict relationship of the classification labels in the material classification direction, analyze the degree of classification offset between classification labels and calculate the degree of conflict in the classification direction, identify abnormal waste, output isolation sorting signal, and obtain abnormal sorting control instructions. The abnormal sorting control instructions include classification direction offset, belonging label conflict level, and sorting path isolation signal.

10. The multimodal waste identification method for intelligent recycling machines based on fuzzy classification according to claim 9, characterized in that, The specific steps for obtaining the abnormal sorting control command are as follows: S501: Call the waste type identification result and the classification level code, determine the matching relationship between the two classification labels in the material classification direction, analyze the classification offset between the labels, calculate the conflict value of the classification direction, and generate classification conflict parameters. S502: Based on the aforementioned classification conflict parameters, identify and filter abnormally classified waste, output an isolation sorting signal and transmit it to the waste sorting equipment, and obtain abnormal sorting control instructions.

Images