An oyster fullness high-throughput non-destructive determination method and system based on ultrasonic waves and deep learning

Abstract

The application discloses a kind of oyster fullness high-throughput nondestructive determination method and system based on ultrasonic wave and deep learning, including sample collection and calibration, the ultrasonic signal of oyster sample and corresponding true fullness index are collected, the digital signal received after penetrating along the long direction of oyster shell and converted is ultrasonic signal, signal pre-processing and feature pattern generation, the digital ultrasonic signal is pre-processed and visual feature pattern is generated;Fullness meets the requirements of goods oyster is sorted to qualified product channel, and oyster with low fullness is sorted to continue fattening and reprocessing channel;The present application uses ultrasonic penetration detection, does not damage oyster at all, can realize online detection of living body, shell, reserves complete raw material for subsequent purification or processing, combined with deep learning model, automatically extract features strongly related to fullness from complex ultrasonic signals, replace artificial experience, realize single determination of second level or even sub-second level.

Description

Technical Field

[0001] This invention relates to the fields of aquaculture and food processing technology, and in particular to a high-throughput non-destructive method and system for measuring oyster plumpness based on ultrasound and deep learning. Background Technology

[0002] Oyster plumpness is a key indicator of their commercial value. Before purification, grading, and processing, quickly and accurately selecting oysters that meet plumpness standards is crucial for improving production efficiency, reducing energy consumption, and increasing product yield. Oyster plumpness determination mainly relies on manual experience or destructive methods. Manual methods are inefficient, subjective, and prone to error; destructive methods cannot be used for online sorting and result in sample waste. Although some studies have attempted to use non-destructive testing techniques such as X-rays and near-infrared spectroscopy, these methods suffer from problems such as expensive equipment, limited penetration, poor adaptability to shell thickness and shape, and the inability to achieve high-speed online detection. Ultrasonic technology, on the other hand, is low-cost, has strong penetration, and is safe and harmless.

[0003] In Chinese patent CN119848598A, a classification model is used to detect oyster plumpness. However, in non-destructive testing of aquatic products, this method is mostly used to determine fish body thickness and fat content. There is no mature solution that applies it to plumpness by linking complex heterogeneous shells with internal soft tissues. In particular, there is a lack of intelligent analysis methods that can adapt to individual differences and high-speed assembly line operations. Therefore, this invention proposes a high-throughput non-destructive testing method and system for oyster plumpness based on ultrasound and deep learning to solve the problems existing in the prior art. Summary of the Invention

[0004] To address the aforementioned problems, the present invention aims to propose a high-throughput non-destructive testing method and system for oyster plumpness based on ultrasound and deep learning. This method and system utilizes ultrasonic penetration detection, which does not damage the oyster at all. It can achieve live and shell-on-site detection, preserving the integrity of the raw material for subsequent purification or processing. Combined with a deep learning model, it automatically extracts features strongly correlated with plumpness from complex ultrasonic signals, replacing manual experience and achieving single-entity measurement at the second or even sub-second level, meeting the needs of industrialized assembly line operations.

[0005] To achieve the objectives of this invention, the invention is implemented through the following technical solution: a high-throughput non-destructive testing method and system for oyster plumpness based on ultrasound and deep learning, comprising the following steps;

[0006] Step 1: Sample collection and calibration. Collect the ultrasonic signal and corresponding true plumpness index of the oyster sample; the ultrasonic signal is a digital signal that is received and converted after penetrating along the length of the oyster shell.

[0007] Step 2: Signal preprocessing and feature graph generation. The digital ultrasonic signal is preprocessed and a visual feature graph is generated.

[0008] Step 3: Deep learning model construction and training. Using the feature image as input and the corresponding true fullness index as label, train the deep learning model to obtain the fullness prediction model.

[0009] Step 4: Online non-destructive testing. For the oyster to be tested, its characteristic image is acquired and input into the plumpness prediction model to obtain its predicted plumpness.

[0010] Step 5: Sorting decision and execution. Based on the predicted plumpness, compare it with the preset plumpness threshold; sort oysters that meet the commercial requirements to the qualified product channel, and sort oysters with low plumpness to the continued fattening and reprocessing channel.

[0011] A further improvement is that, in step one, the ultrasonic wave of a specific frequency is a single frequency and frequency-modulated pulse wave in the range of 200kHz-2MHz, preferably 500kHz-1MHz, to achieve a balance between penetration and resolution.

[0012] A further improvement is that, in step one, the transmitter and receiver are fixed on an adaptively adjustable clamp to ensure that ultrasonic waves can stably penetrate along the length of the shell of oysters of different sizes and shapes.

[0013] A further improvement is that in step two, the visualization is a two-dimensional time-frequency diagram, which retains both the time and frequency domain characteristics of the ultrasonic signal, and more fully reflects the interaction information between the ultrasonic wave and the oyster's internal soft tissue, body fluid and shell structure.

[0014] A further improvement is made in step three, where the deep learning model is a convolutional neural network, preferably a ResNet, DenseNet, or lightweight MobileNet architecture, which adapts to the image input and performs end-to-end feature learning and regression prediction.

[0015] A further improvement is that in step four, the total time for signal acquisition and feature pattern generation is 0.5 seconds.

[0016] A further improvement is made in step five, where the sorting decision is executed by the PLC control system. The system has built-in multi-level fatness threshold parameters: a threshold of ≥0.8 for first-grade products, a threshold of 0.6-0.8 for second-grade products, and a threshold of <0.6 for products to be fattened. These parameters can be customized according to the commodity standards of different breeding enterprises.

[0017] Further improvements include: a conveying and positioning unit, an ultrasonic testing unit, a control and processing unit, and a sorting execution unit;

[0018] Conveying and positioning unit: used to convey oysters in a single, directional, and spaced manner to the inspection station;

[0019] Ultrasonic testing unit: includes ultrasonic transmitting probe, receiving probe, signal generator, and data acquisition card;

[0020] Control and processing unit: including computer and software system, used to control the operation of the entire system, receive digital ultrasonic signals, perform signal preprocessing and feature image generation, run the trained deep learning model, calculate and output the fatness prediction results;

[0021] Sorting execution unit: Based on the instructions issued by the control and processing unit, the oysters are sorted to different outlets, such as qualified product outlet and fattening return outlet.

[0022] A further improvement is that the ultrasonic detection unit includes an adaptively adjustable clamp for fixing the ultrasonic transmitting probe and receiving probe, making it adaptable to oysters of different sizes.

[0023] A further improvement is that the sorting execution unit sorts oysters whose predicted plumpness reaches a preset threshold to the qualified product channel, and sorts oysters that do not meet the standard to the fattening return channel.

[0024] The beneficial effects of this invention are as follows: This invention utilizes ultrasonic penetration detection without damaging the oysters, enabling live and shell-on-site online detection. This preserves the integrity of the raw materials for subsequent purification or processing. Combined with a deep learning model, it automatically extracts features strongly correlated with plumpness from complex ultrasonic signals, replacing manual experience and achieving single-entity measurement at the second or even sub-second level. This meets the needs of industrialized assembly line operations. Through the "graphicalization of ultrasonic signals + deep learning" model, the model can learn the complex mapping relationship between interference factors such as shell thickness, shape, and internal structure and the true plumpness. The measurement results are more objective, accurate, and repeatable than manual experience, directly serving the grading and sorting stage before oyster purification and processing. It can accurately pick out qualified individuals and return unqualified individuals to the fattening stage, significantly improving the yield and consistency of the final product, reducing ineffective processing costs, and improving economic benefits. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0026] Figure 1 This is the front view of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0029] Document CN119848598A describes the detection of oyster plumpness through heterogeneous data collection from the sensor under test and a plumpness classification model. The plumpness classification model training process includes: multi-type data acquisition, data preprocessing, dataset partitioning, data fusion, plumpness classification model training, and model validation. By fusing multi-source heterogeneous data, it provides more comprehensive and complementary information for the plumpness detection of oysters, accurately identifying the plumpness of oysters. It offers advantages such as real-time performance, non-invasiveness, and high accuracy, achieving non-destructive testing of oyster freshness and providing a more powerful non-destructive testing technology for the sustainable development of the oyster industry. However, this application lacks an intelligent analysis method that can adapt to individual differences and high-speed assembly line operations. In this application, the true plumpness index is obtained by dissecting samples, and this is used to train a deep learning model to establish the correlation between the image and plumpness. The trained model is then applied to predict the plumpness of online oysters, and automatic sorting is performed based on the prediction results. The system includes conveying, ultrasonic detection, control processing and sorting units, which realizes rapid, non-destructive, online intelligent detection and sorting of the plumpness of live oysters in their shells.

[0030] Example 1

[0031] according to Figure 1 As shown, this embodiment provides a high-throughput non-destructive testing method and system for oyster plumpness based on ultrasound and deep learning, including the following steps;

[0032] Step 1: Sample collection and calibration. Obtain a batch of representative oyster samples. Using an ultrasonic transmitter of a specific frequency, emit ultrasonic waves along the long direction of the oyster shell (from the top of the shell to the main direction of the ventral edge). After penetrating the entire oyster, the waves are received by a receiver on the opposite side. The receiver converts the ultrasonic signal into an analog electrical signal, which is then converted into a digital ultrasonic signal by a data acquisition card. Simultaneously, the batch of oyster samples is dissected to measure the weight of the soft tissue and the shell, and the plumpness index is calculated as the true label.

[0033] Step Two: Signal Preprocessing and Feature Graphic Generation. The digital ultrasonic signal acquired in Step One is preprocessed, including noise reduction, normalization, and alignment. The processed signal and its frequency domain transformation results (such as Fourier transform and wavelet transform) are used to construct a one-dimensional waveform diagram, a two-dimensional spectrum diagram, and a time-frequency diagram, which serve as visual graphics representing the internal structure of the oyster.

[0034] Step 3: Deep learning model construction and training. Build a deep learning model with the feature image generated in Step 2 as input and the predicted fatness value and fatness level as output. Use the fatness index obtained in Step 1 as training labels, combine the feature image and the corresponding labels to form a training dataset, and train the deep learning model until the model converges to obtain a trained fatness prediction model.

[0035] Step 4: Online non-destructive testing. For oysters with unknown plumpness to be tested, repeat steps 1 and 2 to obtain their feature images. Input the feature images into the plumpness prediction model trained in step 3, and the model will directly output the predicted plumpness value and grade of the oyster.

[0036] Step 5: Sorting decision and execution. Based on the prediction results output in Step 4, compare them with the preset plumpness threshold. Sort oysters that meet the plumpness requirements to the qualified product channel, and sort oysters with low plumpness to the continued fattening and reprocessing channel.

[0037] In step one, the ultrasound of a specific frequency is a single frequency and frequency-modulated pulse wave in the range of 200kHz-2MHz, preferably 500kHz-1MHz, to achieve a balance between penetration and resolution.

[0038] In step one, the transmitter and receiver are fixed on an adaptively adjustable clamp to ensure that ultrasonic waves can stably penetrate along the length of the shell of oysters of different sizes and shapes.

[0039] In step two, the visualization is a two-dimensional time-frequency graph. The graph retains both the time and frequency domain characteristics of the ultrasonic signal, and more fully reflects the interaction information between the ultrasonic wave and the oyster's internal soft tissue, body fluid and shell structure.

[0040] In step three, the deep learning model is a convolutional neural network, preferably a ResNet, DenseNet, or lightweight MobileNet architecture, which adapts to the image input and performs end-to-end feature learning and regression prediction.

[0041] In step four, the total time for signal acquisition and feature pattern generation is 0.5s; meeting the high-throughput detection requirements, it can process ≥1000 samples per hour.

[0042] In step five, the sorting decision is executed by the PLC control system. The system has built-in multi-level plumpness threshold parameters: the threshold for Grade 1 is ≥0.8, the threshold for Grade 2 is 0.6-0.8, and the threshold for oysters to be fattened is <0.6. These parameters can be customized according to the commodity standards of different aquaculture enterprises. The actuator adopts a combination structure of pneumatic push rod and diversion slide. When the oysters pass through the detection station, the PLC controls the pneumatic baffle of the corresponding slide to move according to the predicted level signal, so as to realize the automatic diversion of qualified products and unprocessed products. The response time of the entire sorting process is ≤0.3 seconds, ensuring that it matches the high-throughput detection rhythm.

[0043] The specific waveform of ultrasound is a pulse wave, using short-duration, high-energy pulse signals, which can provide high-resolution data and is suitable for detecting changes in the internal structure and density of oysters. A probe is placed on each side, and the plumpness of the oyster is measured by sending and receiving ultrasound signals. The specific structure of the deep learning model includes a convolutional neural network, an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. Specific parameters include 3-5 convolutional layers, kernel sizes of 3x3 and 5x5, ReLU activation function, an initial learning rate of 0.001, and 100-200 iterations.

[0044] It includes a conveying and positioning unit, an ultrasonic testing unit, a control and processing unit, and a sorting execution unit;

[0045] Conveying and positioning unit: used to convey oysters in a single row, oriented (ensuring consistency along the long direction), and at intervals to the inspection station;

[0046] Ultrasonic testing unit: includes ultrasonic transmitting probe, receiving probe, signal generator, and data acquisition card; the transmitting probe and receiving probe are set opposite each other on both sides of the testing station to transmit and receive ultrasonic signals along the length of the oyster and convert the signals into digital signals;

[0047] Control and processing unit: including computer and software system, used to control the operation of the entire system, receive digital ultrasonic signals, perform signal preprocessing and feature image generation, run the trained deep learning model, calculate and output the fatness prediction results;

[0048] Sorting execution unit: Based on the instructions issued by the control and processing unit, the oysters are sorted to different outlets, such as qualified product outlet and fattening return outlet.

[0049] The ultrasonic testing unit includes an adaptively adjustable clamp for securing the ultrasonic transmitting and receiving probes to fit oysters of different sizes.

[0050] The sorting unit sorts oysters whose predicted plumpness reaches the preset threshold to the qualified product channel, and sorts oysters that do not meet the standard to the fattening return channel.

[0051] Example 2

[0052] Five hundred cultured oysters of varying sizes were selected as a sample. Using a pulsed ultrasonic probe with a center frequency of 800 kHz, and with the support of a custom-designed clamp, the ultrasonic waves were ensured to enter from the top of the shell of each oyster and be received from the ventral margin. Raw waveform data were collected. Each oyster was then dissected, and the fullness index (soft tissue wet weight / total wet weight × 100%) was calculated. Wavelet denoising and amplitude normalization were performed on the raw waveforms. Continuous wavelet transform was then used to convert each one-dimensional signal into a 128x128 pixel two-dimensional time-frequency grayscale image. The time-frequency images and the corresponding fullness index were combined to form a dataset, which was then divided into training, validation, and test sets in a 7:2:1 ratio.

[0053] A lightweight MobileNet-v2 model was used, with the output layer replaced by a regression layer. Mean squared error was used as the loss function, and the Adam optimizer was employed for training. After training, body fatness was predicted on the test set, achieving a determination coefficient R² of 0.91 between the predicted and true values, and a mean absolute error of 0.8%.

[0054] This model was deployed in an online sorting system. Oysters were transported via conveyor belt, and ultrasonic detection was triggered by photoelectric sensors. The system completed signal acquisition, image generation, and model prediction within 0.3 seconds. The plumpness threshold was set at 10%. Oysters with a predicted value ≥10% were pushed into the processing line by pneumatic pushers, while those <10% fell into the return tank to wait for further plumping.

[0055] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A high-throughput non-destructive method for determining oyster plumpness based on ultrasound and deep learning, characterized in that, Includes the following steps; Step 1: Sample collection and calibration. Collect the ultrasonic signal and corresponding true plumpness index of the oyster sample; the ultrasonic signal is a digital signal that is received and converted after penetrating along the length of the oyster shell. Step 2: Signal preprocessing and feature graph generation. The digital ultrasonic signal is preprocessed and a visual feature graph is generated. Step 3: Deep learning model construction and training. Using feature images as input and the corresponding true fullness index as labels, train the deep learning model to obtain the fullness prediction model. Step 4: Online non-destructive testing. For the oyster to be tested, its characteristic image is acquired and input into the plumpness prediction model to obtain its predicted plumpness. Step 5: Sorting decision and execution. Based on the predicted plumpness, compare it with the preset plumpness threshold; sort oysters that meet the plumpness requirements to the qualified product channel, and sort oysters with low plumpness to the continued fattening and reprocessing channel. In step two, the visualization is a two-dimensional time-frequency graph. The graph retains both the time and frequency domain characteristics of the ultrasonic signal, and more fully reflects the interaction information between the ultrasonic wave and the oyster's internal soft tissue, body fluid and shell structure.

2. The high-throughput non-destructive testing method for oyster fullness based on ultrasound and deep learning according to claim 1, characterized in that: In step one, the specific frequency ultrasound is a single frequency and frequency-modulated pulse wave in the range of 200kHz-2MHz.

3. The high-throughput non-destructive testing method for oyster fullness based on ultrasound and deep learning according to claim 1, characterized in that: In step one, the transmitter and receiver are fixed on an adaptively adjustable clamp to ensure that ultrasonic waves can stably penetrate along the length of the shell of oysters of different sizes and shapes.

4. The high-throughput non-destructive testing method for oyster fullness based on ultrasound and deep learning according to claim 2, characterized in that: The specific frequency of the ultrasound is preferably 500kHz-1MHz.

5. The high-throughput non-destructive testing method for oyster fullness based on ultrasound and deep learning according to claim 1, characterized in that: In step three, the deep learning model is a convolutional neural network.

6. The high-throughput non-destructive method for determining oyster plumpness based on ultrasound and deep learning according to claim 5, characterized in that: The convolutional neural network is preferably based on ResNet, DenseNet, or a lightweight MobileNet architecture, which adapts to image input and performs end-to-end feature learning and regression prediction.

7. The high-throughput non-destructive method for determining oyster plumpness based on ultrasound and deep learning according to claim 1, characterized in that: In step four, the total time for signal acquisition and feature pattern generation is 0.5 seconds.

8. The high-throughput non-destructive method for determining oyster plumpness based on ultrasound and deep learning according to claim 1, characterized in that: In step five, the sorting decision is executed by the PLC control system. The system has built-in multi-level fatness threshold parameters: the threshold for first-grade product is ≥0.8, the threshold for second-grade product is 0.6-0.8, and the threshold for product to be fattened is <0.

6. These parameters can be customized according to the commodity standards of different breeding enterprises.

9. A system applied to the high-throughput non-destructive testing method for oyster plumpness based on ultrasound and deep learning as described in any one of claims 1-8, characterized in that: It includes a conveying and positioning unit, an ultrasonic testing unit, a control and processing unit, and a sorting execution unit; Conveying and positioning unit: used to convey oysters in a single, directional, and spaced manner to the inspection station; Ultrasonic testing unit: includes ultrasonic transmitting probe, receiving probe, signal generator, and data acquisition card; Control and processing unit: including computer and software system, used to control the operation of the entire system, receive digital ultrasonic signals, perform signal preprocessing and feature image generation, run the trained deep learning model, calculate and output the fatness prediction results; Sorting execution unit: Based on the instructions issued by the control and processing unit, the oysters are sorted to different outlets, such as qualified product outlet and fattening return outlet.

10. A high-throughput non-destructive method for determining oyster fullness based on ultrasound and deep learning according to claim 9, characterized in that: The ultrasonic testing unit includes an adaptively adjustable clamp for fixing the ultrasonic transmitting probe and receiving probe to accommodate oysters of different sizes.

11. A high-throughput non-destructive method for determining oyster plumpness based on ultrasound and deep learning according to claim 9, characterized in that: The sorting execution unit sorts oysters whose predicted plumpness reaches a preset threshold to the qualified product channel, and sorts oysters that do not meet the standard to the fattening return channel.

Images