An online monitoring method and system for pepper hollow degree and kernel plumpness

Abstract

The application discloses a kind of hollow degree and grain fullness of pepper multi-dimensional acoustic online monitoring method and system, it is related to agricultural product quality detection technical field, method includes: by reticular conveyor belt makes the uniform speed of the pepper monomer to be measured in sound insulation shielding channel movement;Utilize air coupling ultrasonic emission probe to the pepper monomer to be measured directional emission ultrasonic wave, utilize transmission wave receiving probe to capture transmission wave after penetrating the pepper, simultaneously utilize micro sound emission sensor to capture the micro sound emission signal of the pepper;Extract the cavity resonance frequency offset of transmission wave and sound energy attenuation coefficient, carry out band energy extraction to the micro sound emission signal to calculate sound energy entropy;After multi-dimensional acoustic characteristics are standardized respectively, as multi-dimensional feature vector, input into the support vector regression model that is trained in advance, output the hollow and fullness comprehensive evaluation index of the pepper monomer to be measured.This scheme can reduce detection cost, no radiation and can accurately penetrate epidermis to obtain internal grain real state.

Description

Technical Field

[0001] This invention relates to the field of agricultural product quality testing technology, and in particular to a multi-dimensional acoustic online monitoring method and system for the hollowness and seed plumpness of chili peppers. Background Technology

[0002] In the intensive processing of chili peppers (such as high-quality chili oil and specialty hot pot base), the content and plumpness of chili seeds directly determine the fat-soluble aroma, spiciness release stability, and oil quality of the finished product. Therefore, precise screening of dried chili peppers for hollowness and seed plumpness on the production line is a crucial step in increasing product added value. However, existing technologies face significant technical bottlenecks when detecting irregularly hollow characteristics in agricultural products like chili peppers. Dynamic weighing method: Distinguishing between hollow and seeded peppers by weight. However, the thickness and moisture content of pepper skin vary greatly, often resulting in the weight of a thick-skinned, seedless pepper being the same as that of a thin-skinned, seeded pepper, leading to a high misjudgment rate.

[0003] Machine vision and hyperspectral technology: limited by the penetration depth of light waves, it can only extract the texture, color or surface defects of the epidermis, and cannot eliminate blind spots in the internal tissue structure.

[0004] X-ray imaging: Although it can penetrate and form images, industrial-grade food X-ray machines are extremely expensive to build and maintain, and there are compliance risks related to ionizing radiation in conventional agricultural and sideline product processing plants, resulting in low consumer acceptance.

[0005] Currently, the industry urgently needs a low-cost, radiation-free, and non-destructive online detection solution that can accurately penetrate the epidermis to obtain the true state of the internal grains. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a multi-dimensional acoustic online monitoring method and system for chili pepper hollowness and seed plumpness, which can reduce detection costs, is radiation-free, and can accurately penetrate the epidermis to obtain the true state of the internal seeds.

[0007] In order to achieve the objective of this invention, the following solution is proposed: Firstly, a multi-dimensional acoustic online monitoring method for the hollowness and seed plumpness of chili peppers is provided, including the following steps: S1: The individual chili peppers to be tested are moved at a constant speed in the soundproof and shielded channel by a mesh conveyor belt; S2: At the testing station, an air-coupled ultrasonic transmitting probe is used to directionally emit ultrasonic waves to the individual chili pepper to be tested. A transmission wave receiving probe is used to capture the transmitted waves after penetrating the chili pepper. At the same time, a micro-acoustic emission sensor is used to capture the micro-acoustic emission signal of the chili pepper. S3: Perform time-frequency domain analysis on the transmitted wave and the microacoustic emission signal to extract the cavity resonant frequency shift Δ of the transmitted wave.f The acoustic energy attenuation coefficient is used, and wavelet packet decomposition is employed to extract the frequency band energy of the micro-acoustic emission signal in order to calculate the acoustic energy entropy. E e ; S4: Shift the cavity resonant frequency by Δ f Sound energy attenuation coefficient and sound energy entropy E e After standardization, the vectors are used as multidimensional feature vectors and input into a pre-trained support vector regression model to output a comprehensive evaluation index of the hollowness and plumpness of the chili pepper individual.

[0008] Secondly, an online monitoring system for the hollowness and seed fullness of chili peppers is provided to implement the aforementioned multidimensional acoustic online monitoring method, including: The soundproof shielding channel is lined with porous sound-absorbing material, and the outer shell of the channel is made of grounded metal plate; A mesh conveyor belt runs through a soundproof and shielded passage to transport the chili peppers to be tested. An air-coupled ultrasonic transmitting probe and a transmitted wave receiving probe are located opposite each other within a soundproof shielded passage. A low-frequency emission sensor is installed at the top of the soundproof shielding channel; The industrial control processing unit is electrically connected to the air-coupled ultrasonic transmitting probe, the transmitted wave receiving probe, and the micro-acoustic emission sensor, respectively. It is equipped with a pre-trained support vector regression model for processing acoustic features and outputting a comprehensive evaluation index of hollowness and fullness.

[0009] The beneficial effects of this invention are as follows: 1. This invention is the first to apply air-coupled ultrasound and acoustic emission technology in the field of industrial non-destructive testing to the internal density detection of hollow agricultural products. It avoids surface interference from optical vision and radiation risks from X-rays, meets food safety standards, and the air-coupled ultrasound emission probe has a long lifespan and simple structure, with hardware costs several times lower than traditional machine vision, X-ray equipment or CT scanning equipment.

[0010] 2. This invention breaks through the limitation of conventional nondestructive testing relying solely on a single signal, and creatively extracts three physically complementary features simultaneously: the cavity resonant frequency offset Δ, which reflects the size of the macroscopic cavity. f The sound energy attenuation coefficient, which reflects the overall density of the medium, and the sound energy entropy, which reflects the dispersion of micro-particles. E e This greatly improves the detection accuracy of the online monitoring system.

[0011] 3. On the hardware side, the V-shaped limiting groove of the mesh conveyor belt works in conjunction with the laser beam switch to eliminate acoustic acquisition deviation caused by the randomness of material posture and achieve millisecond-level precise triggering.

[0012] 4. The feature extraction process involves minimal computation, achieving millisecond-level response. It directly outputs the comprehensive evaluation index of hollowness and fullness, as well as the classification grade of chili peppers, making it easy to seamlessly integrate with existing automated processing lines.

[0013] 5. Unlike traditional classifiers that directly output rigid good / bad labels, the support vector regression model configured in the industrial control processing unit of this invention can output a continuous numerical comprehensive evaluation index of hollowness and fullness between 0 and 100. Processing enterprises can flexibly fine-tune the threshold boundaries of categories A, B, and C in the software backend according to the quality requirements of different customer orders (for example, raising the threshold for premium fruit to above 90 points). This flexible output mechanism provides strong data support for high-end customization of agricultural product processing. Attached Figure Description

[0014] Figure 1 A flowchart of the online monitoring method is shown; Figure 2 A schematic diagram of the online monitoring system is shown. Detailed Implementation

[0015] Example 1 like Figure 2 As shown in the figure, this embodiment provides a multi-dimensional acoustic online monitoring system for the hollowness and seed plumpness of chili peppers, as detailed below: The main body of the monitoring system is a soundproof shielded tunnel. The tunnel's interior is lined with porous polyurethane sound-absorbing foam to isolate high-frequency mechanical noise from the workshop. The tunnel's outer shell uses a grounded metal plate to shield against external electromagnetic interference. A polytetrafluoroethylene (PTFE) mesh conveyor belt runs through the tunnel to transport the chili peppers. The mesh conveyor belt has an opening ratio greater than 80%. Using a high-opening-ratio PTFE mesh conveyor belt significantly reduces specular reflection interference from the bottom surface. The surface of the mesh conveyor belt has V-shaped limiting grooves to constrain the chili peppers' posture. The opening width and the angle between the V-shaped limiting grooves are pre-adapted to the average radial dimension of the chili peppers after coarse sieving, ensuring that chili peppers of different individual species can lie sideways within the grooves, and that the long axis of the chili peppers is approximately aligned with the conveying direction.

[0016] At the inspection station in the middle of the passage, an air-coupled ultrasonic transmitting probe with a center frequency of 100kHz is installed on one side, with its transmitting end face maintained at a distance of 3 to 5 cm from the chili pepper in the center of the mesh conveyor belt; a transmitted wave receiving probe is installed on the opposite side; a micro-acoustic emission sensor is arranged at the top of the passage, which is connected in series with a high-pass filter with a frequency of 50kHz to filter low-frequency noise such as conveyor belt operation. All probes and micro-acoustic emission sensors are connected to a high-speed data acquisition card with a sampling rate of not less than 1MS / s and connected to the industrial control processing unit.

[0017] A laser beam switch is installed upstream of the air-coupled ultrasonic transmitter, meaning the direction opposite to the chili pepper conveying direction. To prevent the laser beam switch from being accidentally triggered by the V-shaped limiting groove of the mesh conveyor belt itself, the optical axis height of the laser beam switch is configured to be higher than the highest edge of the V-shaped limiting groove and lower than the highest contour of the upper side of the chili pepper being tested in a lateral position.

[0018] The industrial control processing unit is equipped with a pre-trained support vector regression (SVR) model. The industrial control processing unit is used to process acoustic feature data in real time, calculate the comprehensive evaluation index of hollowness and fullness, and then compare the comprehensive evaluation index of hollowness and fullness with the preset threshold range to determine the classification level of the chili pepper: Class A (highly full, no obvious cavity), Class B (partially full, partially hollow), and Class C (hollow, no seeds inside).

[0019] In addition, for high-capacity demands, a multi-channel parallel array design can be adopted, which involves setting up multiple mesh conveyor belts and multiple sets of probes side by side to work simultaneously.

[0020] Example 2 Before the monitoring system of Example 1 is put into production, a support vector regression (SVR) model needs to be constructed, as follows: Step 1: Sample Acquisition and Signal Acquisition Five hundred chili pepper samples were selected and dried and rehydrated to achieve a moisture content of 8%-15%.

[0021] Five hundred chili pepper samples passed through a soundproof shielded channel one by one at a uniform speed. When a single chili pepper sample passed by and triggered the upstream laser beam switch, the industrial control processing unit calculated the delayed trigger time based on the running speed of the mesh conveyor belt and the fixed physical distance between the laser optical axis and the ultrasonic acoustic axis. After the delayed trigger time, the industrial control processing unit controlled the air-coupled ultrasonic transmitting probe to emit ultrasonic waves, ensuring that the single chili pepper sample was running and within the effective sound beam coverage range of the air-coupled ultrasonic transmitting probe. After the air-coupled ultrasonic transmitting probe emitted ultrasonic waves, the transmission wave receiving probe collected and captured the transmitted waves of the chili pepper, and the micro-acoustic emission sensor simultaneously collected and captured the micro-acoustic emission signal of the chili pepper.

[0022] After signal acquisition, 500 samples were destructively dissected. Based on the actual internal state of the peppers, their true hollowness and fullness scores were manually evaluated and used as the continuous training target values ​​for the model.

[0023] In this embodiment, a highly full state is defined as 80 to 100 points, a partially full state as 50 to 80 points, and a hollow state as 0 to 50 points.

[0024] Step 2: Feature Extraction and Physical Mapping (1) The industrial control processing unit extracts the cavity resonant frequency offset Δ of the transmitted wave. f And the sound energy attenuation coefficient. For Class C samples with larger internal cavities, the air medium has a stronger attenuation effect on ultrasound, resulting in a larger sound energy attenuation coefficient. Furthermore, the Helmholtz resonance effect causes a shift in the cavity's resonant frequency Δ. f Significantly increased.

[0025] (2) The industrial control processing unit uses wavelet packet decomposition to extract the frequency band energy of the micro-acoustic emission signal, thereby calculating the acoustic energy entropy. E e , In the formula, n The total number of frequency bands in the wavelet packet decomposition. P i For the first i The proportion of acoustic energy in each frequency band to the total acoustic energy. For Class A samples with denser internal grains, the more intense the grain friction and collision caused by ultrasonic radiation, the more complex the sound wave scattering path, and the corresponding acoustic energy entropy. E e The value is the largest.

[0026] In this embodiment, at a transmission frequency of 100kHz, the acoustic energy attenuation coefficient of the Class C sample was measured to be greater than 15dB / cm, and the cavity resonant frequency shift Δ f Greater than 5.0kHz; while the internally dense Class A sample, due to its good acoustic continuity, has an acoustic energy attenuation coefficient of less than 8dB / cm, almost no resonant frequency shift, and a cavity resonant frequency shift Δ f Less than 1.5kHz. This embodiment measures the Class A acoustic energy entropy. E e Greater than 2.5, Class C sound energy entropy E e Less than 1.2.

[0027] Step 3: Data Standardization Processing Due to the extracted cavity resonance frequency offset Δ f The values ​​and units of the sound energy attenuation coefficient (in kHz), sound energy entropy (in dB / cm), and sound energy entropy (unitless) are completely different. Directly mixing these inputs will cause model bias, and features with large absolute values ​​will mask features with small values. Therefore, this embodiment uses the Z-score method to unify the scale of data from all dimensions. The standardization formula is as follows: z =( x - μ ) / σ ,in x These are the original values ​​of the acoustic features. μ The mean of the acoustic features. σLet be the standard deviation of the acoustic features. After standardization, the three types of features with different dimensions are uniformly converted into dimensionless values ​​with a mean of 0 and a variance of 1, thus constructing a multidimensional feature vector.

[0028] Step 4: Model Building and Classification Decision Output The standardized multidimensional feature vectors are input into the Support Vector Regression (SVR) model for training. Considering the highly nonlinear relationship between acoustic features and chili pepper fullness, this embodiment introduces a radial basis function (RBF). Its function is to nonlinearly map linearly inseparable feature data in the low-dimensional space to the high-dimensional feature space and transform it into linearly separable states, thereby making it easier to construct a decision hyperplane that can accurately separate the three states of "class A, class B, and class C".

[0029] To prevent overfitting during training, a 5-fold cross-validation method was used: 500 sets of data were randomly divided into 5 subsets, with 4 subsets used as the training set and 1 subset as the test set. A grid search method was used to automatically test different parameter combinations and determine two parameters related to model stability: the penalty coefficient. C The kernel function parameter γ is used until the model's prediction error on the test set is lower than a preset threshold.

[0030] In this embodiment, the penalty coefficient C =100, ensuring extremely low tolerance for misclassified samples; kernel function parameter γ=0.01, ensuring that a single extreme sample will not excessively interfere with the overall classification trend.

[0031] After training with the specific parameters described above, the SVR model can directly provide a comprehensive evaluation index of the hollowness and plumpness of individual pepper plants under test. The final classification accuracy of the online monitoring system on the test set is consistently above 95%, and the inference time for a single matrix operation is only 12 milliseconds, fully meeting the real-time requirements of online monitoring.

[0032] Example 3 like Figure 1 As shown in the figure, this embodiment provides an online monitoring method, and the specific workflow is as follows: When the laser beam switch detects a single chili pepper to be tested, the industrial control processing unit controls the air-coupled ultrasonic transmitter to delay the emission of ultrasonic waves. After the air-coupled ultrasonic transmitter emits ultrasonic waves, the transmission wave receiving probe collects and captures the transmitted waves of the chili pepper, and the micro-acoustic emission sensor simultaneously collects and captures the micro-acoustic emission signal of the chili pepper.

[0033] The industrial control processing unit directly performs time-frequency domain analysis on the captured transmitted waves and micro-acoustic emission signals: on the one hand, it performs a fast Fourier transform on the transmitted waves to extract the cavity resonant frequency offset Δ, which reflects the size of the macroscopic cavity.f And the acoustic energy attenuation coefficient, which reflects the overall density of the medium; on the other hand, wavelet packet decomposition is performed on the micro-acoustic emission signal to calculate the acoustic energy entropy, which reflects the discreteness of micro-particles. E e .

[0034] The industrial control processing unit extracts the cavity resonant frequency offset Δ f Sound energy attenuation coefficient and sound energy entropy E e After Z-score standardization, the result is used as a multidimensional feature vector input into a pre-trained support vector regression (SVR) model. After the SVR model performs the calculation, it outputs the comprehensive evaluation index of the chili pepper's hollowness and plumpness in real time.

[0035] Subsequently, the industrial control processing unit compares the comprehensive evaluation index of hollowness and fullness with a preset threshold range to determine the classification grade of the chili pepper (i.e., category A, category B, or category C). If it is determined to be "category C", the downstream pneumatic nozzle is triggered to blow the chili pepper into the defective product box after a preset delay time.

[0036] The above embodiments are only used to illustrate the technical ideas and features of the present invention, and are not intended to be unique or to limit the present invention. Those skilled in the art should understand that various changes or equivalent substitutions made to the present invention without departing from its scope are all within the scope of protection of the present invention.

Claims

1. A multi-dimensional acoustic online monitoring method for the hollowness and seed plumpness of chili peppers, characterized in that, include: S1: The individual chili peppers to be tested are moved at a constant speed in the soundproof and shielded channel by a mesh conveyor belt; S2: At the testing station, an air-coupled ultrasonic transmitting probe is used to directionally emit ultrasonic waves to the individual chili pepper to be tested. A transmission wave receiving probe is used to capture the transmitted waves after penetrating the chili pepper. At the same time, a micro-acoustic emission sensor is used to capture the micro-acoustic emission signal of the chili pepper. S3: Perform time-frequency domain analysis on the transmitted wave and the microacoustic emission signal to extract the cavity resonant frequency shift Δ of the transmitted wave. f The acoustic energy attenuation coefficient is used, and wavelet packet decomposition is employed to extract the frequency band energy of the micro-acoustic emission signal in order to calculate the acoustic energy entropy. E e ; S4: Shift the cavity resonant frequency by Δ f Sound energy attenuation coefficient and sound energy entropy E e After standardization, the vectors are used as multidimensional feature vectors and input into a pre-trained support vector regression model to output a comprehensive evaluation index of the hollowness and plumpness of the chili pepper individual.

2. The multidimensional acoustic online monitoring method according to claim 1, characterized in that, In step S1, the air-coupled ultrasonic transmitter emits ultrasonic waves in a frequency range of 50 kHz to 200 kHz.

3. The multidimensional acoustic online monitoring method according to claim 1, characterized in that, In step S2, a laser beam switch is set up along the conveying direction of the chili pepper to be tested and upstream of the air-coupled ultrasonic transmitting probe. When the chili pepper to be tested triggers the laser beam switch, the delay trigger time is calculated based on the running speed of the mesh conveyor belt and the preset physical distance. After the delay trigger time, the air-coupled ultrasonic transmitting probe emits ultrasonic waves.

4. The multidimensional acoustic online monitoring method according to claim 1, characterized in that, In step S3, the sound energy entropy E e The formula is: In the formula, n The total number of frequency bands in the wavelet packet decomposition. P i For the first i The proportion of sound energy in each frequency band to the total sound energy.

5. The multidimensional acoustic online monitoring method according to claim 1, characterized in that, In step S4, the Z-score method is used to determine the cavity resonant frequency shift Δ. f The sound energy attenuation coefficient and sound energy entropy are standardized separately, and the standardization formula is as follows: z =( x - μ ) / σ In the formula, x These are the original values ​​of the acoustic features. μ The mean of the acoustic features. σ denoted as the standard deviation of the acoustic characteristics.

6. The multidimensional acoustic online monitoring method according to claim 1, characterized in that, In step S4, the training steps for the support vector regression model include: A chili pepper sample set was obtained and processed to achieve a moisture content of 8%-15%. Transmission wave and microacoustic emission signals of each chili pepper sample were collected. After collection, all chili pepper samples were destructively dissected to obtain the true classification labels. Extract the cavity resonant frequency shift Δ of the transmitted wave f The acoustic energy attenuation coefficient is used, and wavelet packet decomposition is employed to extract the frequency band energy of the micro-acoustic emission signal in order to calculate the acoustic energy entropy. E e Construct multidimensional feature vectors; The multidimensional feature vectors and the true classification labels are divided into training and test sets. The support vector regression model is trained using cross-validation, and the penalty coefficient and kernel function parameters are determined by grid search until the prediction error of the model on the test set is lower than the preset threshold.

7. The multidimensional acoustic online monitoring method according to claim 5, characterized in that, In step S4, a radial basis function is introduced when training the support vector regression model to nonlinearly map the linearly inseparable feature data in the low-dimensional space to the high-dimensional feature space and transform it into a linearly separable state.

8. An online monitoring system for the hollowness and seed plumpness of chili peppers, characterized in that, The method for implementing the multidimensional acoustic online monitoring method according to any one of claims 1 to 7 includes: The soundproof shielding channel is lined with porous sound-absorbing material, and the outer shell of the channel is made of grounded metal plate; A mesh conveyor belt runs through a soundproof and shielded channel, and its surface is provided with V-shaped limiting grooves to constrain the posture of the chili peppers; An air-coupled ultrasonic transmitting probe and a transmitted wave receiving probe are located opposite each other within a soundproof shielded passage. A low-frequency emission sensor is installed at the top of the soundproof shielding channel; The industrial control processing unit is electrically connected to the air-coupled ultrasonic transmitting probe, the transmitted wave receiving probe, and the micro-acoustic emission sensor, respectively. It is equipped with a pre-trained support vector regression model for processing acoustic features and outputting a comprehensive evaluation index of hollowness and fullness.

9. The online monitoring system according to claim 7, characterized in that, It also includes a laser beam switch located upstream of the air-coupled ultrasonic transmitter probe.

10. The online monitoring system according to claim 7, characterized in that, The low-acoustic emission sensor is equipped with a 50kHz high-pass filter.

Images