Automatic waste classification method

Abstract

The application relates to the technical field of intelligent environmental protection, and discloses an automatic garbage classification method based on weighing, which comprises the following steps: acquiring image information of a target object to be classified at a drop opening, identifying an initial screening category and calling corresponding standard density and stiffness characteristic ranges; controlling the target object to freely fall onto an elastic weighing assembly at the bottom, exciting mechanical vibration response by using falling impact force; collecting the mechanical vibration response and extracting measured stiffness characteristic values (including main frequency and amplitude attenuation rate), and meanwhile, calculating a measured density value by combining a static weight and an image volume; comparing the measured density value and the measured stiffness characteristic value with the standard ranges, and if both of them are consistent, confirming classification and weighing, otherwise, determining as abnormal. The application realizes accurate verification of internal material and filling state of garbage by using a passive impact principle and through multi-dimensional physical characteristic fusion, and effectively prevents liquid residue and fraud behaviors.

Description

Technical Field

[0001] This invention relates to the field of intelligent environmental protection technology, and in particular to an automatic weighing and sorting method for waste. Background Technology

[0002] With the acceleration of urbanization and the improvement of environmental awareness, intelligent waste sorting and recycling equipment has gradually become widespread. Existing intelligent recycling bins mainly rely on two technological approaches: one is computer vision-based recognition technology, which uses cameras to collect images of waste for sorting; the other is weighing sensor-based measurement technology, which calculates points or amounts based on weight.

[0003] However, existing technologies have significant drawbacks in practical applications. First, simple visual recognition can only capture the external shape features of an object and cannot detect its internal filling. For example, when a user throws in a mineral water bottle filled with water, or a cardboard box filled with bricks, the visual system will still identify it as a "plastic bottle" or "paper," leading to misclassification. Second, simple static weighing cannot distinguish between "effective weight" and "adulterated weight." The aforementioned water-filled bottle will be counted as a high amount of plastic recycling points, which not only causes economic losses for the operator but also leads to a decrease in the purity of the recyclables and increases subsequent processing costs.

[0004] While technologies for material detection using X-rays or active acoustic stimulation (such as tapping devices) exist in the industrial sector, these devices are bulky, expensive, and complex, making them difficult to apply to widely distributed smart trash can terminals in civilian applications. Therefore, there is an urgent need for a low-cost, automated sorting method that requires no additional excitation source and can effectively identify the internal material properties and filling status of trash. Summary of the Invention

[0005] This invention provides an automatic weighing and sorting method for waste. By integrating visual recognition and dynamic vibration feature analysis of drop impact, it achieves non-destructive detection of the internal material and filling state of waste without the need for an additional excitation source, effectively solving the technical problem that existing intelligent recycling equipment cannot identify liquid residues and malicious adulteration.

[0006] This invention provides an automatic weighing and sorting method for waste, comprising:

[0007] S1. Obtain image information of the target object to be classified at the delivery port, use a pre-trained image recognition model to identify the initial screening category of the target object to be classified, and retrieve the corresponding standard density threshold range and standard stiffness feature range from the preset database according to the initial screening category.

[0008] S2. Control the target object to be classified to fall freely from the delivery port to the elastic weighing component at the bottom, and use the impact force generated by the falling target object as a physical excitation to stimulate the elastic weighing component to generate a mechanical vibration response.

[0009] S3. Acquire the mechanical vibration response and convert it into a time-domain vibration signal, perform frequency domain analysis and envelope analysis to extract the measured stiffness feature value; at the same time, after the target object to be classified is stationary, acquire the static weight value, calculate the volume of the target object to be classified based on the image information, and calculate the measured density value in combination with the static weight value; wherein, the measured stiffness feature value includes the dominant frequency and the amplitude attenuation rate.

[0010] S4. Compare the measured density value with the standard density threshold range, and compare the measured stiffness characteristic value with the standard stiffness characteristic range; if both the measured density value and the measured stiffness characteristic value fall within the corresponding standard range, then the initial screening category is confirmed as the final classification result of the target object to be classified and weighing is performed; if either the measured density value or the measured stiffness characteristic value exceeds the corresponding standard range, then the target object to be classified is determined to be abnormal, and an abnormality feedback signal is generated.

[0011] Furthermore, S1 specifically includes:

[0012] S101. The sensor detects that the target object to be classified has entered the delivery area, triggering the visual acquisition module to acquire the original image data of the target object to be classified. The original image data includes color information and depth information. The visual acquisition module uses an RGB-D depth camera or a binocular stereo vision camera.

[0013] S102. Denoising preprocessing is performed on the original image data. Background subtraction method is used to remove the background information of the delivery port in the original image data. The edge contour of the target object to be classified is extracted by combining the Canny operator or Sobel operator to generate pure image data containing only the pixel information of the target object.

[0014] S103. Input the clean image data into the pre-trained image recognition model, output the probability vector of the target object to be classified belonging to each preset category, and select the category with the highest probability value that exceeds the preset confidence threshold as the initial screening category; wherein, the pre-trained image recognition model adopts a convolutional neural network architecture or a Transformer vision architecture.

[0015] S104. Using the initial screening category as an index, retrieve the corresponding standard density threshold range and standard stiffness characteristic range from the preset database; wherein, the standard density threshold range is set based on the material density of the object, and the standard stiffness characteristic range is set based on the structural stiffness and damping characteristics of the object.

[0016] S105. If the probability values ​​of all categories output by the image recognition model are lower than the preset confidence threshold, it is determined that it cannot be recognized, and subsequent steps are not executed and a prompt signal is generated.

[0017] Furthermore, S2 specifically includes:

[0018] S201. After receiving the classification instruction, control the release mechanism at the bottom of the delivery channel to make the target object to be classified start free fall from a preset height away from the elastic weighing component.

[0019] S202, the target object to be classified impacts the elastic weighing component at the bottom, and its falling kinetic energy is converted into impact deformation energy that causes the elastic weighing component to deform and vibration excitation energy that excites vibration.

[0020] S203. The impact force generated when the target object to be classified falls is used as a pulse excitation signal and applied to the elastic weighing component.

[0021] S204. Under the action of the pulse excitation signal, the elastic weighing component is excited to generate a mechanical vibration response including low-frequency rigid body modes and high-frequency structural modes; wherein, the mechanical vibration response includes low-frequency rigid body modes for subsequent static weighing stability, and high-frequency structural modes for material property identification.

[0022] Furthermore, S3 specifically includes:

[0023] S301. Using a sensor array installed at the bottom of the elastic weighing assembly, a time-domain vibration signal characterizing the mechanical vibration response and a static pressure signal characterizing the force change are simultaneously acquired; wherein, the sensor array includes a high-frequency piezoelectric vibration sensor or MEMS accelerometer for acquiring the time-domain vibration signal, and a resistance strain gauge weighing sensor for acquiring the static pressure signal.

[0024] S302. The static pressure signal is low-pass filtered to obtain the static weight value. Based on the image information, the volume of the target object to be classified is calculated using the convex hull algorithm or the minimum bounding box algorithm. The measured density value is calculated by combining the static weight value and the volume.

[0025] S303. Perform high-pass filtering preprocessing on the time-domain vibration signal to remove the low-frequency rigid body swaying component caused by gravity, and extract the effective signal segment.

[0026] S304. Perform a fast Fourier transform on the preprocessed effective signal segment to convert the time-domain signal into a frequency-domain power spectrum, and extract the frequency corresponding to the energy peak as the main frequency.

[0027] S305. Perform Hilbert transform on the preprocessed effective signal segment to extract the instantaneous amplitude envelope, and perform exponential decay fitting on the instantaneous amplitude envelope to calculate the amplitude decay rate.

[0028] S306. The measured density value, main frequency and amplitude attenuation rate are combined into a measured physical characteristic group.

[0029] Furthermore, in step S302, the formula for calculating the measured density value is as follows:

[0030]

[0031] in, This is the static weight value. The volume of the target object to be classified. This is the measured density value.

[0032] Furthermore, in step S305, the amplitude attenuation rate is obtained by least-squares fitting of the instantaneous amplitude envelope, and the fitting formula is:

[0033]

[0034] in, This represents the instantaneous amplitude envelope value at time t; The initial amplitude represents the moment the vibration begins; t represents the time variable. The amplitude attenuation rate is the value to be determined, which represents the internal damping characteristics and energy dissipation rate of the target object to be classified.

[0035] Furthermore, S4 specifically includes:

[0036] S401. The central processing unit reads the measured physical feature group and, using the initial screening category as an index, loads the corresponding standard density threshold range and standard stiffness feature range from the preset database.

[0037] S402. When the measured density value falls within the range of the standard density threshold, a density compliance flag is generated.

[0038] S403. Determine whether the main frequency and amplitude attenuation rate fall within the corresponding main frequency range and attenuation rate range of the standard stiffness characteristic range, respectively. Only when both fall within the corresponding range, generate a stiffness compliance flag.

[0039] S404. Perform a logical AND operation on the density compliance flag and the stiffness compliance flag. If the operation result is true, the classification is determined to be accurate and without fraud, and the initial screening category is determined to be the final classification result; if the operation result is false, the classification is determined to be abnormal.

[0040] S405. Based on the determination result of step S404, if the classification is determined to be accurate, control the sorting flap below the elastic weighing component to open, unload the target object to be classified into the storage area of ​​the corresponding category, and accumulate the static weight value to the user account; if the classification is determined to be abnormal, control the voice feedback module to broadcast the reason for the abnormality, and control the mechanical structure of the delivery channel to perform locking or retraction actions to refuse to accept the target object to be classified.

[0041] Furthermore, S403 specifically includes:

[0042] Determine whether the measured main frequency falls within the standard main frequency range. If so, it indicates that the material hardness of the target object is consistent with the initial screening category.

[0043] Determine whether the measured amplitude attenuation rate falls within the standard attenuation rate range. If so, it indicates that the internal damping characteristics of the target object are normal. Only when the above two sub-judgments are satisfied simultaneously, a stiffness compliance flag is generated.

[0044] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0045] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0046] The beneficial effects of this invention are as follows:

[0047] This invention cleverly converts gravitational potential energy into mechanical vibration response by utilizing the natural impact force of falling waste during disposal as a passive physical excitation. This eliminates the need for additional active excitation devices (such as hammers or radiation sources), significantly reducing hardware costs and energy consumption. By fusing volume information obtained through visual recognition with the dynamic vibration characteristics (dominant frequency and amplitude attenuation rate) generated by the impact, this invention can not only calculate the measured density but also deeply analyze the stiffness and internal damping characteristics of the object. This multimodal fusion detection mechanism can sensitively identify the physical differences between empty and full bottles, and between pure materials and adulterated substances, effectively addressing the shortcomings of existing technologies in detecting liquid residues and malicious weight gain fraud. This significantly improves the accuracy of waste sorting and the cleanliness of recyclables. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating the automatic weighing and sorting method for waste according to the present invention.

[0049] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0050] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0051] like Figure 1 As shown, the present invention provides an automatic weighing and sorting method for waste, comprising:

[0052] S1. Obtain image information of the target object to be classified at the delivery port, use a pre-trained image recognition model to identify the initial screening category of the target object to be classified, and retrieve the corresponding standard density threshold range and standard stiffness feature range from the preset database according to the initial screening category.

[0053] In one embodiment, step S1 specifically includes the following sub-steps:

[0054] S101. When a user approaches the smart trash can or opens the disposal opening baffle, an infrared distance sensor or photoelectric switch installed on the top inner side of the disposal opening detects the object entering. At this time, the visual acquisition module located above the disposal channel is activated. The visual acquisition module (preferably an RGB-D depth camera or a binocular camera), in conjunction with auxiliary lighting (such as a high color rendering LED fill light), captures images of the target object to be classified in the disposal area, acquiring raw image data. This raw image data includes the RGB color image information and depth information of the target object to be classified.

[0055] S102. After receiving the raw image data, the central processing unit first performs preprocessing operations, including noise reduction, white balance correction, and brightness enhancement, to eliminate interference from changes in ambient light. Subsequently, using background subtraction or edge detection algorithms (such as the Canny operator), the contour region (ROI) of the target object to be classified is segmented from the image background to remove interference information from the background of the delivery port, obtaining clean image data containing only the target object.

[0056] S103. Input the clean image data obtained in step S102 into the pre-trained image recognition model. In this embodiment, the image recognition model adopts an improved convolutional neural network architecture (such as ResNet-50 or YOLOv8-tiny). This model has been trained on a massive garbage sample dataset (containing garbage images of different angles, deformations, and degrees of dirt) in the cloud or locally. The model outputs a probability vector indicating whether the target object to be classified belongs to each preset category. The category with the highest probability value that exceeds a preset confidence threshold (e.g., 85%) is selected as the initial screening category (e.g., PET plastic bottles, aluminum cans, cardboard boxes, or glass bottles).

[0057] S104. The system internally stores a standard physical property database, which stores the standard physical parameter ranges for various types of waste in key-value pairs. Once the initial screening category is determined, it is immediately used as an index to retrieve the corresponding standard physical fingerprint range from the database, including the standard density threshold range and the standard stiffness characteristic range. Specifically:

[0058] Standard density threshold range ( (This is based on the density of common materials in this category. For example, if the initial screening category is PET plastic bottles, the standard density range should be 0.9~1.4 g / cm³.) 3 (Considering empty bottles or small amounts of residue). If subsequent calculations show that the measured density far exceeds this range (e.g., reaching 3.0 g / cm³),... 3 If the symbol is '('), it suggests that there may be a foreign object inside.'

[0059] Standard stiffness characteristic range ( ): Includes the main frequency range ( ) and amplitude attenuation range ( For example, if the initial screening category is aluminum can (metal), its standard main frequency is usually high (e.g., >1000Hz, crisp sound) and the attenuation rate is extremely low (the vibration lasts a long time); if the initial screening category is paper ball, its standard main frequency is low and messy, and the attenuation rate is extremely high (the sound is dull, and the energy is absorbed quickly).

[0060] S105. If, in step S103, the probability of all categories output by the image recognition model is lower than the confidence threshold (i.e., it cannot identify what the object is), or the recognition result belongs to a non-acceptable category (such as being identified as a brick, a large metal part, etc.), skip the subsequent steps directly, control the voice module to prompt that it cannot be recognized or prohibit delivery, and keep the delivery channel closed.

[0061] S2. Control the target object to be classified to fall freely from the delivery port onto the elastic weighing component at the bottom, and use the impact force generated by the falling target object as a physical excitation to stimulate the elastic weighing component to generate a mechanical vibration response.

[0062] In one embodiment, step S2 utilizes the physical principle of converting gravitational potential energy into kinetic energy, and specifically includes the following sub-steps:

[0063] S201. After image acquisition and initial screening are completed in step S1, if the object is determined to be of an acceptable category, the central control unit sends an opening command to the electromagnetic release mechanism located below the delivery channel. The electromagnetic release mechanism controls the baffle (or flap) at the bottom of the delivery channel to open instantly. At this time, the object to be classified, under the action of gravity, begins to fall freely from a preset height H (e.g., 40cm~60cm) above the bottom elastic weighing component. Setting a fixed drop height H is to ensure that different batches of tests have a consistent initial kinetic energy benchmark. This ensures the comparability of subsequent vibration analyses.

[0064] S202. The object to be classified falls vertically and impacts the elastic weighing component at the bottom. The elastic weighing component includes a rigid bearing plate, an elastic support (such as a high-strength spring or damping rubber block), and a rigid base. At the moment of contact, the kinetic energy of the object to be classified is rapidly converted into two parts of energy:

[0065] ① Impact deformation energy: causes compressive deformation of the elastic support component;

[0066] ② Vibration excitation energy: Excites the entire bearing plate to generate high-frequency vibration.

[0067] S203. The impact force generated by the drop acts as a pulse excitation signal on the elastic weighing component. This impact force is not uniformly distributed, but exhibits different transmission characteristics depending on the landing posture and material hardness of the target object to be classified.

[0068] If the target object is a rigid body (such as a glass bottle), the impact force has a short duration and a large peak value, producing a near-ideal Dirac effect. Function excitation;

[0069] If the target object is a soft object (such as a plastic bag full of garbage), the impact force has a long duration, and the peak value is weakened by the buffering effect, resulting in a relatively gentle pulse excitation.

[0070] S204. Under the aforementioned physical excitation, the elastic weighing component generates a mechanical vibration response. This response comprises two superimposed physical motion modes:

[0071] ①Low-frequency rigid body mode: The entire bearing plate undergoes low-frequency up-and-down swaying on the elastic support (usually 5Hz~20Hz), and this motion will eventually stabilize for static weighing.

[0072] ② High-frequency structural mode: High-frequency vibration and sound wave transmission (usually 100Hz~5000Hz) generated by impact on the surface and internal medium of the bearing disk.

[0073] The mechanical vibration response described in step S4 specifically refers to the composite physical motion containing the aforementioned high-frequency and low-frequency components, providing a physical source for the sensor signal acquisition in subsequent step S3.

[0074] S3. Acquire the mechanical vibration response and convert it into a time-domain vibration signal, perform frequency domain analysis and envelope analysis to extract the measured stiffness feature value; at the same time, after the target object to be classified is stationary, acquire the static weight value, calculate the volume of the target object to be classified based on the image information, and calculate the measured density value in combination with the static weight value; wherein, the measured stiffness feature value includes the dominant frequency and the amplitude attenuation rate.

[0075] In one embodiment, step S3 deconstructs the physical essence of the target object to be classified through in-depth analysis of the raw sensor data, employing a parallel processing architecture to simultaneously process static data (for density calculation) and dynamic data (for stiffness feature calculation). Specifically, it includes the following sub-steps:

[0076] S301. During the mechanical vibration response described in step S2, the sensor array below the elastic weighing assembly begins to operate.

[0077] Dynamic channel: A high-frequency piezoelectric vibration sensor (or MEMS accelerometer) mounted on the bottom of the support plate continuously acquires voltage signals at a high sampling rate (e.g., 10kHz~44.1kHz), converting the mechanical vibration response into a digital time-domain vibration signal. .

[0078] Static channel: A resistance strain gauge load cell installed below the elastic support synchronously acquires pressure change signals.

[0079] S302. After a brief period of oscillation on the support plate, the target object to be classified tends to come to a stop (usually within 0.5 to 1 second). The signal from the resistance strain gauge load cell is then low-pass filtered (cutoff frequency set to, for example, 5Hz) to remove high-frequency jitter interference and obtain a stable static weight value. Simultaneously, the image information obtained in step S1 is invoked. A 3D point cloud of the target object to be classified is constructed based on RGB-D depth data, and the volume of the target object to be classified is calculated using either the convex hull algorithm or the minimum bounding box algorithm. .

[0080] Finally, according to the formula The measured density value was calculated.

[0081] S303, The time-domain vibration signal obtained in step S301 Preprocessing is performed. First, a high-pass filter (cutoff frequency set to 20Hz, for example) is used to remove the low-frequency rigid body sway components caused by gravity, retaining only the high-frequency structural vibration components determined solely by the impact material properties. Second, an endpoint detection algorithm is used to extract effective signal segments from the moment of impact until the vibration energy decays to the background noise level, which serve as samples for subsequent analysis.

[0082] S304. Perform a Fast Fourier Transform (FFT) on the preprocessed effective signal segment to convert the time-domain signal into a frequency-domain power spectrum. , specifically:

[0083] The frequency corresponding to the peak point with the highest energy in the frequency domain power spectrum is the main frequency. . The modulus of a material is positively correlated with its Young's modulus; this applies to high-hardness materials such as metals and glass. Typically higher (crisp), while lower-hardness materials such as plastic and paper... Low (dull).

[0084] S305. Process the preprocessed time-domain signal Perform a Hilbert transform to extract the instantaneous amplitude envelope of the signal. The least squares method is used to analyze the envelope. Perform exponential decay fitting, the fitting function model is as follows: ,in That is, the amplitude attenuation rate to be determined. This reflects the energy dissipation capacity within an object; an empty bottle (single medium) decays slowly; a bottle filled with liquid (fluid-structure interaction) or a paper bag filled with wet waste, due to its high internal damping, has its energy rapidly absorbed, leading to... Significantly increased.

[0085] S306. Package the parameters calculated above into a measured physical characteristic group:

[0086]

[0087] The feature set will be transmitted to the central processing unit for the final comparison and determination in step S4.

[0088] S4. Compare the measured density value with the standard density threshold range, and compare the measured stiffness characteristic value with the standard stiffness characteristic range; if both the measured density value and the measured stiffness characteristic value fall within the corresponding standard range, then the initial screening category is confirmed as the final classification result of the target object to be classified and weighing is performed; if either the measured density value or the measured stiffness characteristic value exceeds the corresponding standard range, then the target object to be classified is determined to be abnormal, and an abnormality feedback signal is generated.

[0089] In one embodiment, step S4 specifically includes the following sub-steps:

[0090] S401, The Central Processing Unit (CPU) establishes the current decision task context; on the one hand, it reads the measured physical feature group from the output of step S3. On the other hand, the initial screening category determined in step S1 (e.g., category ID: C01-PET bottle) is invoked, and the standard physical fingerprint range corresponding to that category is loaded from the database, including:

[0091] Standard density range: Standard frequency range: Standard attenuation rate range: .

[0092] S402, First determine the measured density value Does it fall within the standard density range? Execute the following logical judgment: .

[0093] If the condition is met, a density compliance flag is generated, marking Flag_Density=TRUE. This indicates that the macroscopic mass / volume ratio of the target object conforms to the general physical characteristics of that category (e.g., there is no extreme density deviation such as foam plastic masquerading as metal or an empty bottle filled with mercury).

[0094] If the condition is not met, mark Flag_Density=FALSE.

[0095] S403. Further verification of dynamic features, which includes two parallel or serial sub-decisions:

[0096] ① Hardness matching determination: Determine the measured main frequency Does it fall within the standard clock speed range? If it falls into the range, it means that the material hardness (rigidity) of the target object is consistent with the initial screening category (for example, confirming that it is a crisp sound of metal, rather than a dull sound of stone).

[0097] ② Determining the state of the contents: Judging the measured amplitude attenuation rate Does it fall within the standard attenuation rate range? If the target falls within the range, it indicates that the internal damping characteristics of the target object are normal (e.g., confirming slow decay of an empty bottle, rather than rapid decay of a full bottle of liquid). Only when both of the above sub-judgments are satisfied simultaneously is a stiffness compliance flag generated, marking Flag_Stiffness=TRUE; otherwise, Flag_Stiffness=FALSE is marked.

[0098] S404. Perform a logical AND operation on the two flag bits mentioned above, and output the final instruction based on the result:

[0099] Scenario 1 (Judgment Passed): When (Flag_Density==TRUE)AND(Flag_Stiffness==TRUE): the distribution is judged to be "accurately classified and without fraud". The central processing unit locks the initial screening category as the final classification result, records the static weight value, and adds it to the user's environmental points account.

[0100] Scenario 2 (Abnormal Detection): When (Flag_Density==FALSE)OR(Flag_Stiffness==FALSE): An "abnormality" is determined in this delivery (including misdelivery, malicious fraud, or failure to clear contents). The central processing unit generates an abnormality feedback signal and records the abnormality type code (e.g., Error_01: Density too high, Error_02: Material too soft).

[0101] S405. Based on the decision result of step S404, drive the underlying hardware to perform the corresponding action, specifically:

[0102] If the judgment is successful: control the flip motor below the elastic weighing component to unload the target object into the corresponding category's internal storage bin, and announce "Deployment successful, weight XX grams, points +N" through the display screen or voice.

[0103] If an anomaly is detected: the voice module will announce the specific reason for the anomaly (e.g., liquid detected in the bottle, please empty it before disposal or do not mix heavy objects into the plastic recycling bin); the disposal port baffle will remain locked or the target object will be pushed out to the retrieval port by the reverse push rod, and the target object will be refused to be accepted.

[0104] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0105] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. An automatic weighing and waste sorting method, characterized in that, include: S1. Obtain image information of the target object to be classified at the delivery port, use a pre-trained image recognition model to identify the initial screening category of the target object to be classified, and retrieve the corresponding standard density threshold range and standard stiffness feature range from the preset database according to the initial screening category. S2. Control the target object to be classified to fall freely from the delivery port to the elastic weighing component at the bottom, and use the impact force generated by the falling target object as a physical excitation to stimulate the elastic weighing component to generate a mechanical vibration response. S3. Acquire the mechanical vibration response and convert it into a time-domain vibration signal, perform frequency domain analysis and envelope analysis to extract the measured stiffness feature value; at the same time, after the target object to be classified is stationary, acquire the static weight value, calculate the volume of the target object to be classified based on the image information, and calculate the measured density value in combination with the static weight value; wherein, the measured stiffness feature value includes the dominant frequency and the amplitude attenuation rate. S4. Compare the measured density value with the standard density threshold range, and compare the measured stiffness characteristic value with the standard stiffness characteristic range; if both the measured density value and the measured stiffness characteristic value fall within the corresponding standard range, then the initial screening category is confirmed as the final classification result of the target object to be classified and weighing is performed; if either the measured density value or the measured stiffness characteristic value exceeds the corresponding standard range, then the target object to be classified is determined to be abnormal, and an abnormality feedback signal is generated.

2. The automatic weighing and sorting method for waste according to claim 1, characterized in that, S1 specifically includes: S101. The sensor detects that the target object to be classified has entered the delivery area, triggering the visual acquisition module to acquire the original image data of the target object to be classified. The original image data includes color information and depth information. The visual acquisition module uses an RGB-D depth camera or a binocular stereo vision camera. S102. Denoising preprocessing is performed on the original image data. Background subtraction method is used to remove the background information of the delivery port in the original image data. The edge contour of the target object to be classified is extracted by combining the Canny operator or Sobel operator to generate pure image data containing only the pixel information of the target object. S103. Input the clean image data into the pre-trained image recognition model, output the probability vector of the target object to be classified belonging to each preset category, and select the category with the highest probability value that exceeds the preset confidence threshold as the initial screening category; wherein, the pre-trained image recognition model adopts a convolutional neural network architecture or a Transformer vision architecture. S104. Using the initial screening category as an index, retrieve the corresponding standard density threshold range and standard stiffness characteristic range from the preset database; wherein, the standard density threshold range is set based on the material density of the object, and the standard stiffness characteristic range is set based on the structural stiffness and damping characteristics of the object. S105. If the probability values ​​of all categories output by the image recognition model are lower than the preset confidence threshold, it is determined that it cannot be recognized, and subsequent steps are not executed and a prompt signal is generated.

3. The automatic weighing and sorting method for waste according to claim 1, characterized in that, S2 specifically includes: S201. After receiving the classification instruction, control the release mechanism at the bottom of the delivery channel to make the target object to be classified start free fall from a preset height away from the elastic weighing component. S202, the target object to be classified impacts the elastic weighing component at the bottom, and its falling kinetic energy is converted into impact deformation energy that causes the elastic weighing component to deform and vibration excitation energy that excites vibration. S203. The impact force generated when the target object to be classified falls is used as a pulse excitation signal and applied to the elastic weighing component. S204. Under the action of the pulse excitation signal, the elastic weighing component is excited to generate a mechanical vibration response including low-frequency rigid body modes and high-frequency structural modes; wherein, the mechanical vibration response includes low-frequency rigid body modes for subsequent static weighing stability, and high-frequency structural modes for material property identification.

4. The automatic weighing and sorting method for waste according to claim 1, characterized in that, S3 specifically includes: S301. Using a sensor array installed at the bottom of the elastic weighing assembly, a time-domain vibration signal characterizing the mechanical vibration response and a static pressure signal characterizing the force change are simultaneously acquired; wherein, the sensor array includes a high-frequency piezoelectric vibration sensor or MEMS accelerometer for acquiring the time-domain vibration signal, and a resistance strain gauge weighing sensor for acquiring the static pressure signal. S302. The static pressure signal is low-pass filtered to obtain the static weight value. Based on the image information, the volume of the target object to be classified is calculated using the convex hull algorithm or the minimum bounding box algorithm. The measured density value is calculated by combining the static weight value and the volume. S303. Perform high-pass filtering preprocessing on the time-domain vibration signal to remove the low-frequency rigid body swaying component caused by gravity, and extract the effective signal segment. S304. Perform a fast Fourier transform on the preprocessed effective signal segment to convert the time-domain signal into a frequency-domain power spectrum, and extract the frequency corresponding to the energy peak as the main frequency. S305. Perform Hilbert transform on the preprocessed effective signal segment to extract the instantaneous amplitude envelope, and perform exponential decay fitting on the instantaneous amplitude envelope to calculate the amplitude decay rate. S306. The measured density value, main frequency and amplitude attenuation rate are combined into a measured physical characteristic group.

5. The automatic weighing and sorting method for waste according to claim 4, characterized in that, In step S302, the formula for calculating the measured density value is as follows: in, This is the static weight value. The volume of the target object to be classified. This is the measured density value.

6. The automatic weighing and sorting method for waste according to claim 4, characterized in that, In step S305, the amplitude attenuation rate is obtained by least-squares fitting of the instantaneous amplitude envelope, and the fitting formula is as follows: in, This represents the instantaneous amplitude envelope value at time t; The initial amplitude represents the moment the vibration begins; t represents the time variable. The amplitude attenuation rate is the value to be determined, which represents the internal damping characteristics and energy dissipation rate of the target object to be classified.

7. The automatic weighing and sorting method for waste according to claim 4, characterized in that, S4 specifically includes: S401. The central processing unit reads the measured physical feature group and, using the initial screening category as an index, loads the corresponding standard density threshold range and standard stiffness feature range from the preset database. S402. When the measured density value falls within the range of the standard density threshold, a density compliance flag is generated. S403. Determine whether the main frequency and amplitude attenuation rate fall within the corresponding main frequency range and attenuation rate range of the standard stiffness characteristic range, respectively. Only when both fall within the corresponding range, generate a stiffness compliance flag. S404. Perform a logical AND operation on the density compliance flag and the stiffness compliance flag. If the operation result is true, the classification is determined to be accurate and without fraud, and the initial screening category is determined to be the final classification result; if the operation result is false, the classification is determined to be abnormal. S405. Based on the determination result of step S404, if the classification is determined to be accurate, control the sorting flap below the elastic weighing component to open, unload the target object to be classified into the storage area of ​​the corresponding category, and accumulate the static weight value to the user account; if the classification is determined to be abnormal, control the voice feedback module to broadcast the reason for the abnormality, and control the mechanical structure of the delivery channel to perform locking or retraction actions to refuse to accept the target object to be classified.

8. The automatic weighing and sorting method for waste according to claim 7, characterized in that, Specifically, S403 includes: Determine whether the measured main frequency falls within the standard main frequency range. If so, it indicates that the material hardness of the target object is consistent with the initial screening category. Determine whether the measured amplitude attenuation rate falls within the standard attenuation rate range. If so, it indicates that the internal damping characteristics of the target object are normal. Only when the above two sub-judgments are satisfied simultaneously, a stiffness compliance flag is generated.

Images