Fruit quality damage-free detection and sorting device

Abstract

The invention belongs to the technical field of fruit grading, and provides a fruit quality damage-free detecting and sorting device which comprises a lead screw sliding table, an image sorting module, a sorting and detecting module, a sorting and detecting acquisition circuit and a wireless impedance sensing upper computer. Through the image sorting module and the sorting and detecting module, coupling of visual-dielectric characteristics is achieved, appearance characteristics are considered, electrical properties such as the internal structure and the moisture content are also considered, and more comprehensive and accurate grading can be achieved.

Description

Technical Field

[0001] The invention relates to the technical field of fruit grading, and in particular to a device for non-destructive detection and sorting of fruit quality. Background Art

[0002] Fruits are usually graded multiple times on sorting lines using manual and mechanical devices, which leads to inaccurate and inefficient grading, and it is difficult to judge pesticide residues by the naked eye. Summary of the invention

[0003] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a non-destructive detection and sorting device for fruit quality, which solves the problems of inaccurate and inefficient grading and difficulty in judging pesticide residues caused by multiple grading by manual and mechanical devices on the sorting line.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A non-destructive detection and sorting device for fruit quality, comprising: a screw slide, an image sorting module, a sorting and detection module, a sorting and detection acquisition circuit and a wireless impedance sensor host computer;

[0006] The screw slide is arranged vertically on the ground; the image sorting module is arranged on the top of the screw slide; the sorting and detection module is arranged on the slide of the screw slide; the sorting and detection acquisition circuit is connected to the sorting and detection module; the wireless impedance sensor host computer is connected to the sorting and detection acquisition circuit;

[0007] The image sorting module is used to collect the appearance image of the target fruit; the sorting and detection module is used to collect the bioelectric signal of the target fruit; the sorting and detection acquisition circuit is used to convert the bioelectric signal into a standard acquisition signal; the wireless impedance sensing host computer is used to perform impedance scanning on the standard acquisition signal to obtain impedance amplitude data and phase angle data, and calculate the pesticide residue concentration and classification result of the target fruit based on the impedance amplitude data and phase angle data.

[0008] Preferably, the image sorting module is 320 mm away from the bottom of the screw slide.

[0009] Preferably, the sorting and detection module comprises: a 3D printed slider, an adaptive flexible sheet and a multi-modal flexible sorting and detection sensor; the multi-modal flexible sorting and detection sensor comprises: a flexible impedance sensor and a flexible electrochemical non-enzymatic sensor;

[0010] The flexible impedance sensor and the flexible electrochemical non-enzymatic sensor are symmetrically arranged on a pre-prepared PDMS substrate; the flexible impedance sensor is attached to the surface of the adaptive flexible sheet; the adaptive flexible sheet is adhered to the 3D printed slider; the 3D printed slider is fixed on the slide of the screw slide;

[0011] The flexible impedance sensor and the flexible electrochemical non-enzymatic sensor are used to detect the impedance characteristics and pesticide residue concentration of the target fruit respectively.

[0012] Preferably, the 3D printed slider and the adaptive flexible sheet are both made of elastic materials.

[0013] Preferably, the manufacturing process of the electrode of the flexible impedance sensor includes:

[0014] In an environment of a growth temperature of 1000° C., a single-layer graphene film is grown on the surface of a copper foil by using a mixed gas of methane and hydrogen in a ratio of 1:10 through chemical vapor deposition technology; the growth time of the graphene film is 30 minutes;

[0015] Through the thermal transfer process, 5N / cm 2 The graphene film is transferred to a polydimethylsiloxane substrate having a thickness of 100 μm formed by spin coating with a pressure of ;

[0016] A spiral conductive pattern is constructed on the graphene film by using a dilute nitric acid solution through photolithography and wet etching technology; the spiral conductive pattern has a pitch of 500 μm; and the line width of the spiral conductive pattern is 50 μm;

[0017] The graphene film packaging leads after the conductive pattern construction are packaged with silver nanoparticle conductive adhesive and cured at 120° C. for 10 minutes to obtain the electrodes of the flexible impedance sensor.

[0018] Preferably, the manufacturing process of the flexible electrochemical non-enzymatic sensor comprises:

[0019] A micrometer-scale serpentine resistive conductive electrode pattern was constructed on a pre-prepared polydimethylsiloxane substrate by photolithography and plasma etching technology, and electrode deposition was performed using a composite conductive material of gold and carbon nanotubes;

[0020] The deposited micrometer-scale serpentine resistive conductive electrode pattern is electrochemically pretreated for 300 seconds using a 0.5 mol / L sulfuric acid solution by a constant potential method to obtain a working electrode; the potential range of the electrochemical pretreatment is -2V to +2V;

[0021] 5.0 μL of a composite solution of polyaniline and graphene oxide with a mass concentration of 1 mg / mL was dripped onto the surface of the working electrode and dried at room temperature for 2 hours to obtain the flexible electrochemical non-enzymatic sensor.

[0022] Preferably, the expression of the pesticide residue concentration is:

[0023]

[0024] Among them, C AI is the pesticide residue concentration predicted by the AI ​​model; p is the peak current value of the sample; I b is the background blank current value; v is the scanning rate; E range is the potential scanning range; T is the ambient temperature; f ML (·) is the mapping function based on the machine learning model; X i is the i-th group of inputs to the mapping function; ∈ is the model error term; and n is the total number of inputs to the model.

[0025] Preferably, the classification result expression is:

[0026] in,

[0027] L final is the classification result; is the electrical impedance amplitude; is the electrical impedance phase angle; Represent the impedance signals of yellow fruit and mixed color fruit respectively; is the impedance signal of green fruit; η is the normalized comprehensive quality score; ω j is the weight coefficient; is the hierarchical mapping result; round(·) is the mapping function; n is the number of levels.

[0028] Preferably, the sorting and detection acquisition circuit is embedded with a main control module, a signal conditioning module, an impedance conversion module, a power supply and level conversion module, a key module and a Bluetooth communication module;

[0029] The signal conditioning module amplifies, filters and linearizes the collected original signal; the impedance conversion module is used to convert the bioelectric signal into a voltage signal.

[0030] Preferably, the use process of the wireless impedance sensor host computer includes:

[0031] Setting impedance test parameters and writing the impedance test parameters into a preset register; the impedance test parameters include: frequency range and scanning step length;

[0032] Switch the preset impedance conversion chip to the preheating mode;

[0033] The target fruit is scanned step by step at a preset frequency by the impedance conversion chip, and the data is transmitted to the memory of the single-chip computer in real time by DMA;

[0034] Performing validity check on the collected data in the memory of the single chip microcomputer;

[0035] When the scanning is finished, the impedance conversion chip is switched to a low power consumption mode, and all the collected data that have passed the validity check are sent to a computer terminal.

[0036] The present invention discloses the following technical effects:

[0037] The present invention provides a non-destructive detection and sorting device for fruit quality. Through an image sorting module and a sorting and detection module, the present invention solves the problems of grading inaccuracy and low efficiency as well as difficulty in judging pesticide residues caused by multiple grading by manual and mechanical devices on a sorting line, thereby realizing the coupling of visual-dielectric features. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0039] Figure 1 A schematic diagram of the structure of a non-destructive fruit quality detection and sorting device provided in an embodiment of the present invention;

[0040] Figure 2 Three views of the sorting and detection module provided in the embodiment of the present invention;

[0041] Figure 3 A schematic diagram of a multi-modal flexible sorting and detection sensor provided in an embodiment of the present invention;

[0042] Figure 4 A network structure diagram of YOLOv5s_CBAM provided in an embodiment of the present invention;

[0043] Figure 5 This is a schematic diagram of the evaluation of the YOLOv5s_CBAM model for non-destructive detection and sorting of fruit quality provided by an embodiment of the present invention. Figure 5 (a) is the training loss graph, Figure 5 (b) is the evaluation index diagram, Figure 5 (c) is the confusion matrix, Figure 5(d) is the precision-recall curve;

[0044] Figure 6 A flow chart of non-destructive detection and sorting of fruit quality provided by an embodiment of the present invention;

[0045] Description of reference numerals:

[0046] 1-screw slide, 2-sorting and detection module, 21-3D printing slider, 22-adaptive flexible sheet, 3-image sorting module. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] The purpose of the present invention is to provide a non-destructive detection and sorting device for fruit quality, so as to solve the problems of inaccurate and inefficient grading and difficulty in judging pesticide residues caused by multiple grading by manual and mechanical devices on a sorting line.

[0049] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] Figure 1 The structure diagram of the non-destructive detection and sorting device for fruit quality provided by the embodiment of the present invention is as follows: Figure 1 As shown, the present invention provides a non-destructive detection and sorting device for fruit quality, comprising: a screw slide 1, an image sorting module 3, a sorting and detection module 2, a sorting and detection acquisition circuit and a wireless impedance sensor host computer;

[0051] The screw slide 1 is arranged vertically on the ground; the image sorting module 3 is arranged on the top of the screw slide 1; the sorting and detection module 2 is arranged on the slide of the screw slide 1; the sorting and detection acquisition circuit is connected to the sorting and detection module 2; the wireless impedance sensor host computer is connected to the sorting and detection acquisition circuit;

[0052] The image sorting module 3 is used to collect the appearance image of the target fruit; the sorting and detection module 2 is used to collect the bioelectric signal of the target fruit; the sorting and detection acquisition circuit is used to convert the bioelectric signal into a standard acquisition signal; the wireless impedance sensing host computer is used to perform impedance scanning on the standard acquisition signal to obtain impedance amplitude data and phase angle data, and calculate the pesticide residue concentration and classification result of the target fruit based on the impedance amplitude data and phase angle data.

[0053] Preferably, the image sorting module 3 is 320 mm away from the bottom of the screw slide 1 .

[0054] refer to Figure 2 , the sorting and detection module 2 includes: a 3D printed slider 21, an adaptive flexible sheet 22 and a multi-modal flexible sorting and detection sensor; the multi-modal flexible sorting and detection sensor includes: a flexible impedance sensor and a flexible electrochemical non-enzymatic sensor;

[0055] The flexible impedance sensor and the flexible electrochemical non-enzymatic sensor are symmetrically arranged on a pre-prepared PDMS substrate; the flexible impedance sensor is attached to the surface of the adaptive flexible sheet 22; the adaptive flexible sheet 22 is adhered to the 3D printed slider 21; the 3D printed slider 21 is fixed on the slide of the screw slide 1;

[0056] Flexible impedance sensors and flexible electrochemical non-enzymatic sensors were used to detect the impedance characteristics and pesticide residue concentration of target fruits, respectively.

[0057] Preferably, the 3D printed slider 21 and the adaptive flexible sheet 22 are both made of elastic materials.

[0058] refer to Figure 3 , the manufacturing process of the electrode of the flexible impedance sensor includes:

[0059] Under a growth temperature of 1000°C, a single-layer graphene film was grown on the surface of the copper foil using a mixed gas of methane and hydrogen in a ratio of 1:10 by chemical vapor deposition technology; the growth time of the graphene film was 30 minutes;

[0060] Through the thermal transfer process, 5N / cm 2 The graphene film was transferred to a polydimethylsiloxane substrate with a thickness of 100 μm formed by spin coating with a pressure of 100 μm.

[0061] A spiral conductive pattern is constructed on the graphene film using a diluted nitric acid solution through photolithography and wet etching. The pitch of the spiral conductive pattern is 500 μm. The line width of the spiral conductive pattern is 50 μm.

[0062] The graphene film with the conductive pattern constructed was packaged with silver nanoparticle conductive adhesive and cured at 120°C for 10 minutes to obtain the electrodes of the flexible impedance sensor.

[0063] Preferably, the manufacturing process of the flexible electrochemical non-enzymatic sensor comprises:

[0064] A micrometer-scale serpentine resistive conductive electrode pattern was constructed on a pre-prepared polydimethylsiloxane substrate by photolithography and plasma etching technology, and electrode deposition was performed using a composite conductive material of gold and carbon nanotubes;

[0065] The deposited micrometer-scale serpentine resistive conductive electrode pattern was electrochemically pretreated for 300 seconds using a 0.5 mol / L sulfuric acid solution by a constant potential method to obtain a working electrode; the potential range of the electrochemical pretreatment was -2 V to +2 V;

[0066] 5.0 μL of a composite solution of polyaniline and graphene oxide with a mass concentration of 1 mg / mL was dropped onto the surface of the working electrode and dried at room temperature for 2 hours to obtain a flexible electrochemical non-enzymatic sensor.

[0067] Furthermore, the expression of pesticide residue concentration is:

[0068]

[0069] Among them, C AI is the pesticide residue concentration predicted by the AI ​​model; I p is the peak current value of the sample; I b is the background blank current value; v is the scanning rate; E range is the potential scanning range; T is the ambient temperature; f ML (·) is the mapping function based on the machine learning model; X i is the i-th group of inputs to the mapping function; ∈ is the model error term; n is the total number of inputs to the model.

[0070] Specifically, the classification result expression is:

[0071] in,

[0072] L final is the classification result; is the electrical impedance amplitude; is the electrical impedance phase angle; Represent the impedance signals of yellow fruit and mixed color fruit respectively; is the impedance signal of green fruit; η is the normalized comprehensive quality score, which is calculated by weighted nonlinear operation and has a value range of [0, 1]; ω j is the weight coefficient, which is used to control the influence of different parameters; is the hierarchical mapping result; round(·) is the mapping function; n is the number of levels.

[0073] Optionally, the sorting and detection acquisition circuit is embedded with a main control module, a signal conditioning module, an impedance conversion module, a power supply and level conversion module, a key module and a Bluetooth communication module;

[0074] The signal conditioning module amplifies, filters and linearizes the collected original signal; the impedance conversion module is used to convert the bioelectric signal into a voltage signal.

[0075] Specifically, the use process of the wireless impedance sensor host computer includes:

[0076] Set impedance test parameters and write the impedance test parameters into a preset register; the impedance test parameters include: frequency range and scanning step length;

[0077] Switch the preset impedance conversion chip to the preheating mode;

[0078] The target fruit is scanned step by step at a preset frequency through an impedance conversion chip, and the data is transmitted to the microcontroller memory in real time through DMA;

[0079] Verify the validity of the collected data in the microcontroller memory;

[0080] When the scan is completed, the impedance conversion chip is switched to a low power consumption mode, and all collected data that have passed the validity check are sent to a computer terminal.

[0081] Preferably, the image sorting module 3 uses a USB industrial camera (model HM2131) installed on the top of the screw slide 1, and is connected to the PC to transmit and store images. In addition, in order to capture a complete image, the camera module needs to be at a certain distance from the horizontal plane. After many experiments, it was found that when the camera is 320 mm away from the bottom of the screw slide 1 and the image resolution is 1280×720, the captured image best meets the requirements.

[0082] Optionally, when testing the device-level system, mango is selected as the object. According to the characteristic changes of mango in different periods, it is divided into the following four categories: C grade, the peel is green, the texture is hard, and the flavor has not yet fully developed; M grade, the peel is mixed color, has developed relatively fully or not fully developed; R grade, black and brown spots appear, the texture is soft and rotten, and it is obviously corrupt; Y grade, the peel is bright or too dark in color, the texture is soft, and the flavor is mellow.

[0083] Preferably, after all images are collected, a deep learning model will be used for the initial classification of mangoes. In order to select the most suitable deep learning model, the classification effects of the following different models are compared: YOLOv5s_CBAM model (reference Figure 4 ), YOLOv5s model, YOLOv4 model and other fruit classification models. The same dataset was used to train YOLOV4 and YOLOv5s. The results show that the map value of the YOLOv5s_CBAM model used in this embodiment is 5.32% and 3.11% higher than that of the YOLOv4 and YOLOv5s models, respectively, and its comprehensive performance is also higher than other technologies. Figure 4 , the YOLOv5s_CBAM model network structure combines convolutional neural network (CNN) and attention mechanism (CBAM) and is designed for image detection tasks. The network first performs multi-scale pooling on the input image through the SPPF (Spatial Pyramid Pooling) module to extract feature information of different scales. Then, multiple convolutional layers (Conv) are used for feature extraction, and the C3_CBAM module is embedded in it, combining the convolution operation and the CBAM attention mechanism to enhance the model's attention to important features. The network further restores the spatial resolution through the upsampling layer (Upsample) and uses the concatenation layer (Concat) to merge the feature maps of different layers to enhance the feature expression ability. The following BottleNeck_CBM module further optimizes feature selection through the bottleneck structure and attention mechanism, reduces the amount of calculation and improves network efficiency. The ConBNSiLU module combines convolution, batch normalization and SiLU activation functions to improve the nonlinear expression ability of the network. In order to enhance the learning ability of the network, the ADD module implements cross-layer jump connection through addition operation, which improves the information flow and feature fusion ability. Finally, the network further optimizes the feature map through the CBAM module and finally outputs the detection result. The entire network structure integrates a variety of advanced technologies, and improves the performance and efficiency of image detection tasks through the combination of deep feature extraction and attention mechanism. It is worth noting that YOLOv5s_CBAM2 successfully balances the two key indicators of weight and accuracy, and its weight is only 12% of YOLOV4. Its compact size of 14.5MB and fast image processing time of 9ms emphasize its suitability as a lightweight model for mobile deployment, which helps to integrate it into mobile devices. The CBAM can be expressed as:

[0084]

[0085] Indicates. Where H represents the feature map, W c , W s denote channel-based and spatial-based attention, respectively. represents element-by-element multiplication, H” and H′ represent the output feature maps after channel attention and spatial attention, respectively.

[0086] Furthermore, Python 3.9.17 was used for programming, the deep learning framework was Pytorch, and the training hardware was a computer with Intel i9 13900HX CPU, 32G RAM and NVIDIA GeForce RTX 4090 32G. The Adam optimizer was used in the training, the learning rate was set to 0.001, the MiniBatchSize was set to 2, and the MaxEpochs was set to 10. This is how the first classification of mangoes can begin.

[0087] Specifically, after the first visual classification, Figure 5 (a) Observe the changes in the model's loss during training, including training loss and validation loss (box_loss, obj_loss, cls_loss). As the training progresses, these loss functions basically converge after 150 iterations. This shows that the model has successfully learned to locate and identify mangoes and determine whether the target exists, achieving a stable training effect. Figure 5 (b) It can be seen that the detection indicators of specific mango categories in the test set performed well. For different mango categories (Cyan, Mix, Rotten, Yellow), the model's precision (Precision), recall (Recall) and average precision (mAP) all showed high levels. When the IoU threshold was set to 0.5, the average precision (mAP) of the model reached 0.960; within the stricter IoU threshold range (0.5-0.95), the mAP value was 0.781. This shows that the model not only performs well under loose conditions, but also maintains a high classification ability even under stricter conditions, meeting the detection requirements of mango classification application scenarios. Attached Figure 5 (c) shows the confusion matrix of the model, which is used to evaluate the model's classification ability for different categories. The confusion matrix shows the distribution of predicted values ​​and true values, where the values ​​on the diagonal represent the proportion of correct classifications. It can be seen that the model's prediction performance for most categories is relatively accurate. For example, the prediction accuracy of the Cyan category is 0.81, and the accuracy of the Mix category is as high as 0.96, indicating that the model has a strong ability to distinguish between these two types of mangoes. At the same time, the accuracy of the Rotten category is 0.95 and the Yellow category is 0.94, which also reflects the reliability of the model in identifying these categories. Although there are a small number of misclassifications (such as the Cyan category being partially classified as the Yellow category), the overall impact is small and the overall classification ability of the model is good.

[0088] Further, according to the confusion matrix generated by the test set, the model performed well on the M, R, and Y classes with the highest accuracy, and the classification accuracy of the C class was relatively low. This difference may be due to the smoothness of the transition from unripe to ripe mangoes, which makes the model face greater challenges in classifying this grade. Overall, the model is more inclined to predict samples as more mature grades, which indicates that the model has a stricter classification standard. AP value of each mango category and PR attachment of the model Figure 5(d) shows the precision-recall curves of different mango categories to further verify the model performance. The curves show that the precision of the Cyan category is 0.938, the Mix category is 0.980, the Rotten category is 0.975, and the Yellow category is 0.948. Overall, the overall mAP of all categories is 0.960, which reflects the high accuracy and stability of the model in the classification task. In addition, the figure also provides representative sample images of each category, which provides a more intuitive explanation of the classification results and further illustrates the reliability and interpretability of the model in practical applications.

[0089] Optionally, the size of the 3D printed slider 21 and the adaptive flexible sheet 22 is designed and processed according to the size of the slide table and the flexible impedance electrode used. In this embodiment, the base size of the 3D printed slider 21 is 50 mm×80 mm.

[0090] Preferably, the adaptive flexible sheet 22 and the 3D printed slider 21 are both made of elastic material, so that the sensor can be tightly attached to the surface of the mango without damaging the mango, and the impedance signal can be better collected.

[0091] Preferably, the flexible impedance electrode of the multimodal flexible sorting and detection sensor consists of an electrode layer, a substrate layer, and a lead layer. During its operation, the graphene electrode is tightly attached to the surface of the fruit when subjected to external force. The PDMS substrate serves as a flexible dielectric, and the electrode collects the impedance signal of the fruit skin and transmits it to the back-end processing module. When the maximum transverse and longitudinal stretching rate of the flexible impedance electrode is 40%, the electrode resistance changes by less than 3%, and can recover to the initial resistance after 200 stretching cycles. The spiral design effectively improves the stretchability and stability of the electrode, ensuring its accuracy and reliability in fruit grading and detection, and is suitable for the non-destructive detection needs of fruits with complex curved surfaces.

[0092] Specifically, the flexible impedance electrode transfers a serpentine resistive conductive electrode onto a silicone substrate. The resistance of the electrode of this shape can quickly recover to its initial state after high-frequency and multiple horizontal and vertical stretching. The electrode after the stretching cycle can meet the requirements of normal operation during mango grading.

[0093] Preferably, this embodiment uses a rigid annular impedance electrode with a commercial electrochemical workstation to test the real impedance value of mangoes. A total of 700 groups of data were obtained in the experiment, including 148 groups of C, 339 groups of M, and 213 groups of Y. The three mangoes C, M, and Y were also divided into 6 categories using a rigid impedance sensing system: Y1 grade, overripe, too dark in color, over-softened; Y2 grade, fully mature, bright yellow peel, soft texture, best edible state; M1 grade, mature, mixed color peel, soft texture, and more uniform maturity; M2 grade, uneven maturity, mixed color peel, and some areas are mature; C1 grade, immature, green or light green peel, hard texture; C2 grade, immature, green peel, and hard texture.

[0094] Specifically, the modified electrode is connected to the flexible sorting and detection collection circuit at the back end through a flexible wire, and the pesticide residue signal is detected using electrochemical impedance spectroscopy (EIS) technology, and the pesticide concentration in the fruit skin is quantitatively analyzed in combination with a data analysis algorithm. The concentration formula is:

[0095]

[0096] After 1000 continuous mechanical bending cycles, the anodic peak current in the cyclic voltammogram of the RF / pretreated electrode did not change significantly, confirming the flexibility of the flexible electrochemical non-enzymatic sensor.

[0097] Preferably, when the flexible sorting and detection module 2 is working, the flexible impedance electrode is attached to the bottom of the adaptive flexible sheet 22, and the lead screw slide is controlled to descend, so that the pressure applied to the flexible sensor will make it tightly attached to the mango. In this process, the signal received by the impedance electrode is transmitted to the back-end PCB circuit for further processing. The simulation of the impedance measurement process can be divided into three steps. First, the impedance signal of each mango is collected based on the test steps. The impedance signal of each mango includes amplitude and phase angle, and the calculation formula is as follows:

[0098]

[0099] Where Z is R d The collection of impedance characteristic data in Z Ae and Z Pe Denote amplitude samples and phase angle samples, respectively. I denotes current and R denotes resistance. Secondly, mangoes with feedback signals in the same range are classified into the same category. Each of the original three categories (yellow, mixed color, and cyan) is further divided into two categories. The calculation formula is as follows:

[0100]

[0101] in, as well as They represent the impedance signals of the first category yellow mangoes, mixed color mangoes and green mangoes and the impedance signals of the second category yellow mangoes, mixed color mangoes and green mangoes respectively. Finally, the signals of the same category of mangoes are averaged and compared with the average value of the signals of another category of mangoes in the same category. The threshold coefficient is set according to the average value of the impedance signals of the two categories of mangoes, and the classification of mangoes is finally completed. The calculation formula is as follows:

[0102]

[0103] in, The electrical impedance amplitude indicates the internal structure, sugar content and moisture uniformity of the fruit. The electrical impedance phase angle reflects the internal electrical properties and maturity of the fruit. The impedance signals of yellow fruit and mixed color fruit are respectively used to judge the color characteristics and quality differences of fruits. Impedance signal of green fruit, through nonlinear compression Avoid excessively large values ​​that may adversely affect the classification results. final : Final classification result. After the flexible electrochemical non-enzymatic sensor comes into contact with the surface of the fruit, an electrochemical reaction occurs. The flexible sorting and detection acquisition circuit at the back end obtains the electrochemical signal, and then uses the obtained data to draw the corresponding voltammetric curve to analyze the concentration of specific pesticides remaining on the skin.

[0104] Specifically, the flexible sorting and detection acquisition circuit is mainly composed of the main control module, which is responsible for the control and coordination of the entire sorting and detection system, including receiving sensor signals, controlling actuator actions, data processing and decision-making. The signal conditioning module is mainly used to amplify, filter and linearize the original signal collected by the sensor to improve the signal-to-noise ratio and accuracy, and adapt to the input requirements of the main control module. The impedance conversion module is used to convert the bioelectric signal collected by the electrochemical sensor into a voltage signal, which is usually achieved through impedance matching networks, amplifiers and other circuits. The power supply and level conversion module is used to provide stable power supply for each module in the system and perform level conversion on the signal. The key module provides a human-computer interaction method. The user can start, stop, set parameters and other operations on the system through key operations, thereby realizing the control of the system. And the Bluetooth communication module is used to transmit the data collected by the system to external devices through Bluetooth wireless. In addition, the conversion module includes an impedance conversion chip, and the frequency generator inside the chip can generate excitation signals of different frequencies. The excitation signal is output through the output electrode of the FISS, and after passing through the sample to be tested, the input signal is received by the input electrode of the FISS. The DSP module inside the conversion module performs discrete Fourier transform processing on the signal, and then transmits the processed signal to the main control board. The USART serial port inside the main control board is responsible for transmitting the signal to the BT-11 Bluetooth module to achieve wireless data transmission. The entire system is powered by a 5V lithium battery. Finally, the flexible sorting and detection acquisition circuit at the back end is connected to the multi-modal flexible sorting and detection sensor to obtain the impedance signal of the fruit.

[0105] Specifically, the functions of the wireless impedance sensor host computer mainly include real-time acquisition and display of impedance parameters, and the output file format is csv. When clicking to start the test, 0x550xaa is sent via Bluetooth, and the single chip will assign "testing" to 1. During the test, click to stop the test or press the button, send 0xaa0x55, the single chip will execute "testing = 0", and then interrupt the test. In addition, the control program of the impedance sensing system simultaneously collects the amplitude and phase angle of mango impedance. First, the system sets the impedance test parameters (such as frequency range, scanning step, etc.) through the host computer and writes these parameters into the configuration register. Subsequently, the impedance conversion chip switches to the preheating mode to ensure that the chip automatically starts the impedance scan after reaching a stable working state. During the scanning process, the chip gradually excites the fruit skin at a preset frequency, collects the corresponding impedance amplitude and phase angle data, and transmits the data to the single chip memory in real time through DMA (direct memory access). After receiving the complete data packet, the single chip microcomputer verifies the validity of the data (such as removing noise and outliers). When the entire frequency scan is completed, the chip automatically enters low-power mode, and the microcontroller packages the collected data and sends it to the host computer via Bluetooth to complete the current task. After the secondary classification, the percentage of correct classification can be seen, which is the classification accuracy of the flexible impedance system. In this process, 13 groups of mangoes out of 700 groups of mangoes were not correctly classified, which may be caused by the uneven maturity of mangoes. The accuracy of mango secondary classification is as high as 98.14%, which meets the needs of mango grading on the sorting line and proves that this classification method is feasible.

[0106] refer to Figure 6 , the overall detection process of this embodiment is as follows: In this system, the fruit and vegetable samples to be detected are first placed at the designated sensor position for the first preprocessing. The system starts to collect signals from the surface of fruits and vegetables, and performs preliminary processing through sensors and converts them into a data format suitable for analysis. Next, the system uses fast Fourier transform (FFT) to convert time domain signals into frequency domain signals, performs spectrum analysis, and extracts frequency features. Then, the frequency data is impedance analyzed with the help of the AD5933 module to ensure the accuracy and quality of the data. Through further processing of the signal, the system determines the skin color and depth of fruits and vegetables, which helps to evaluate their freshness. If the skin color of fruits and vegetables meets the preset standards, the system will decide whether to assign them to the next processing stage. Finally, after classification and judgment, the system outputs the processing results and analysis data through the data transmission module, uses machine learning technology to classify and process the data in real time, generates a final classification report, and stores the test results.

[0107] The beneficial effects of the present invention are as follows:

[0108] The present invention proposes a non-destructive detection and sorting device and technical method for fruit quality, which introduces dielectric impedance features as compensation for image sorting. Dielectric impedance is an important indicator of electrical properties, which can reflect information such as the internal structure and water content of an object. By analyzing the dielectric impedance data of the fruit, a self-made multi-modal flexible sorting and detection sensor is used to perform secondary non-destructive sorting of the fruit; through the coupling of vision-dielectric features-electrochemical properties, not only the features on the appearance are taken into account, but also the electrical properties such as the internal structure and water content, as well as the pesticide residues on the surface, which can achieve more comprehensive and accurate sorting and detection. It provides a method for the comprehensive utilization of multi-modal features in the agricultural field, which is expected to promote the development of product grading in agricultural production and can be promoted for other fruits.

[0109] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0110] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A non-destructive detection and sorting device for fruit quality, characterized in that: include: Screw slide, image sorting module, sorting and detection module, sorting and detection acquisition circuit and wireless impedance sensor host computer; The screw slide is arranged vertically on the ground; the image sorting module is arranged on the top of the screw slide; the sorting and detection module is arranged on the slide of the screw slide; the sorting and detection acquisition circuit is connected to the sorting and detection module; the wireless impedance sensor host computer is connected to the sorting and detection acquisition circuit; The image sorting module is used to collect the appearance image of the target fruit; the sorting and detection module is used to collect the bioelectric signal of the target fruit; the sorting and detection acquisition circuit is used to convert the bioelectric signal into a standard acquisition signal; the wireless impedance sensing host computer is used to perform impedance scanning on the standard acquisition signal to obtain impedance amplitude data and phase angle data, and calculate the pesticide residue concentration and classification result of the target fruit based on the impedance amplitude data and phase angle data.

2. The device for non-destructive detection and sorting of fruit quality according to claim 1, characterized in that: The image sorting module is 320 mm away from the bottom of the screw slide.

3. The device for non-destructive detection and sorting of fruit quality according to claim 1, characterized in that: The sorting and detection module includes: a 3D printed slider, an adaptive flexible sheet, and a multi-modal flexible sorting and detection sensor; the multi-modal flexible sorting and detection sensor includes: a flexible impedance sensor and a flexible electrochemical non-enzymatic sensor; The flexible impedance sensor and the flexible electrochemical non-enzymatic sensor are symmetrically arranged on a pre-prepared PDMS substrate; the flexible impedance sensor is attached to the surface of the adaptive flexible sheet; the adaptive flexible sheet is adhered to the 3D printed slider; the 3D printed slider is fixed on the slide of the screw slide; The flexible impedance sensor and the flexible electrochemical non-enzymatic sensor are used to detect the impedance characteristics and pesticide residue concentration of the target fruit respectively.

4. The device for non-destructive detection and sorting of fruit quality according to claim 3, characterized in that: The 3D printed slider and the adaptive flexible sheet are both made of elastic materials.

5. The device for non-destructive detection and sorting of fruit quality according to claim 3, characterized in that: The manufacturing process of the electrode of the flexible impedance sensor includes: In an environment of a growth temperature of 1000° C., a single-layer graphene film is grown on the surface of a copper foil by using a mixed gas of methane and hydrogen in a ratio of 1:10 through chemical vapor deposition technology; the growth time of the graphene film is 30 minutes; Through the thermal transfer process, 5N / cm 2 The graphene film is transferred to a polydimethylsiloxane substrate having a thickness of 100 μm formed by spin coating with a pressure of ; A spiral conductive pattern is constructed on the graphene film by using a dilute nitric acid solution through photolithography and wet etching technology; the spiral conductive pattern has a pitch of 500 μm; and the line width of the spiral conductive pattern is 50 μm; The graphene film packaging leads after the conductive pattern construction are packaged with silver nanoparticle conductive adhesive and cured at 120° C. for 10 minutes to obtain the electrodes of the flexible impedance sensor.

6. The device for non-destructive detection and sorting of fruit quality according to claim 3, characterized in that: The manufacturing process of the flexible electrochemical non-enzymatic sensor comprises: A micrometer-scale serpentine resistive conductive electrode pattern was constructed on a pre-prepared polydimethylsiloxane substrate by photolithography and plasma etching technology, and electrode deposition was performed using a composite conductive material of gold and carbon nanotubes; The deposited micrometer-scale serpentine resistive conductive electrode pattern is electrochemically pretreated for 300 seconds using a 0.5 mol / L sulfuric acid solution by a constant potential method to obtain a working electrode; the potential range of the electrochemical pretreatment is -2V to +2V; 5.0 μL of a composite solution of polyaniline and graphene oxide with a mass concentration of 1 mg / mL was dripped onto the surface of the working electrode and dried at room temperature for 2 hours to obtain the flexible electrochemical non-enzymatic sensor.

7. The device for non-destructive detection and sorting of fruit quality according to claim 1, characterized in that: The expression of the pesticide residue concentration is: Among them, C AI is the pesticide residue concentration predicted by the AI ​​model; p is the peak current value of the sample; I b is the background blank current value; v is the scanning rate; E range is the potential scanning range; T is the ambient temperature; f ML (·) is the mapping function based on the machine learning model; X i is the i-th group of inputs to the mapping function; ∈ is the model error term; and n is the total number of inputs to the model.

8. The device for non-destructive detection and sorting of fruit quality according to claim 1, characterized in that: The classification result expression is: in, L final is the classification result; is the electrical impedance amplitude; is the electrical impedance phase angle; Represent the impedance signals of yellow fruit and mixed color fruit respectively; is the impedance signal of green fruit; η is the normalized comprehensive quality score; ω j is the weight coefficient; is the hierarchical mapping result; round(·) is the mapping function; n is the number of levels.

9. The device for non-destructive detection and sorting of fruit quality according to claim 1, characterized in that: The sorting and detection acquisition circuit is embedded with a main control module, a signal conditioning module, an impedance conversion module, a power supply and level conversion module, a key module and a Bluetooth communication module; The signal conditioning module amplifies, filters and linearizes the collected original signal; the impedance conversion module is used to convert the bioelectric signal into a voltage signal.

10. The device for non-destructive detection and sorting of fruit quality according to claim 1, characterized in that: The use process of the wireless impedance sensor host computer includes: Setting impedance test parameters and writing the impedance test parameters into a preset register; the impedance test parameters include: frequency range and scanning step length; Switch the preset impedance conversion chip to the preheating mode; The target fruit is scanned step by step at a preset frequency by the impedance conversion chip, and the data is transmitted to the memory of the single-chip computer in real time by DMA; Performing validity check on the collected data in the memory of the single chip microcomputer; When the scanning is finished, the impedance conversion chip is switched to a low power consumption mode, and all the collected data that have passed the validity check are sent to a computer terminal.

Images