Mutton freshness nondestructive bionic detection equipment and method

Abstract

The invention provides nondestructive bionic detection equipment and method for mutton freshness, and relates to the technical field of flexible sensing, and the equipment comprises a flexible sensor array, a transverse transmission rod, a longitudinal transmission rod sleeve, a viscosity detector and other components. The sensor is connected with the fixing device through the guide rail, so that the mutton freshness is accurately detected. The device is flexible and accurate, and the state of mutton can be comprehensively detected through cooperative work of various sensors such as a pressure sensor and a mechanical vision sensor. The operation is simple and convenient, and the non-destructive detection method ensures the product quality and food safety.

Description

Technical Field

[0001] The invention relates to the field of flexible sensing technology, and in particular to a non-destructive bionic detection device and method for mutton freshness. Background Art

[0002] Meat is susceptible to the effects of enzymes, microorganisms, etc. during storage, transportation, and processing, causing corruption and deterioration, resulting in a decline in quality. The supply and demand sides of mutton are geographically far away and rely heavily on imports, so the freshness of mutton has also attracted the attention of consumers and regulators. According to investigations and readings of literature, the surface characteristics of meat are closely related to the freshness of meat, and the existing mechanical vision detection for evaluating the surface characteristics of mutton is not very versatile. Therefore, it is necessary to improve the recognition of the surface characteristics of mutton to achieve a more accurate freshness prediction. However, domestic research in this direction is not in-depth. The emergence of multimodal data fusion technology provides a new method for extracting surface features. Summary of the invention

[0003] In order to overcome the deficiencies of the prior art, the object of the present invention is to provide a non-destructive bionic detection device and method for mutton freshness.

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

[0005] A non-destructive bionic detection device for mutton freshness, comprising: a flexible sensor array, a transverse transmission rod, a longitudinal transmission rod sleeve, a longitudinal transmission rod top, a transverse axis and longitudinal axis converter, a viscous detector guide rail fixing device, a pressure detector guide rail fixing device, a viscous detector integrated device, a viscous probe sleeve, a flexible strain sensor, a pressure sensor probe, a flexible pressure sensor, a mechanical vision sensor, a stage, a box, a door, a control panel and a host computer;

[0006] The viscous probe cover is connected to the viscous detector integrated device through an axial moving pair; the viscous detector integrated device is fixed below the viscous detector guide rail fixing device through threads; the transverse transmission rod passes through the entire viscous detector guide rail fixing device through two bearings of the viscous detector guide rail fixing device;

[0007] The pressure sensor probe is connected to the lower part of the pressure detector guide rail fixing device through an axial moving pair; the transverse transmission rod passes through the pressure detector guide rail fixing device through the bearing of the pressure detector guide rail fixing device; the transverse axis longitudinal axis converter is connected to one end of the transverse transmission rod; the longitudinal transmission rod sleeve is sleeved on the shaft hole below the transverse axis longitudinal axis converter; the longitudinal transmission rod top is connected to the longitudinal transmission rod sleeve through a thread; the longitudinal transmission rod top is fixed at a preset position of the box; the stage is connected to the box with screws; the mechanical vision sensor is fixed to the corresponding position of the box; the flexible sensor array is fixed to the stage by inlaying; the door is connected to the box through a door shaft, and the control panel is installed on the box; the flexible strain sensor is pasted on the working surface below the adhesive probe cover by pasting; the flexible pressure sensor is pasted on the working surface below the pressure sensor probe by pasting; the host computer is respectively connected to the flexible pressure sensor, the mechanical vision sensor and the flexible strain sensor; the flexible sensor array is used to obtain a pressure change signal reflecting the change of pressure value.

[0008] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0009] The present invention provides a non-destructive bionic detection device and method for the freshness of mutton. The device comprises: a flexible sensor array, a transverse transmission rod, a longitudinal transmission rod sleeve, a longitudinal transmission rod top, a transverse axis and longitudinal axis converter, a viscous detector guide rail fixing device, a pressure detector guide rail fixing device, a viscous detector integrated device, a viscous probe sleeve, a flexible strain sensor, a pressure sensor probe, a flexible pressure sensor, a mechanical vision sensor, a stage, a box, a door, a control panel and a host computer; the viscous probe sleeve is connected to the viscous detector integrated device through an axial moving pair; the viscous detector integrated device is fixed to the bottom of the viscous detector guide rail fixing device through a thread; the transverse transmission rod passes through the entire viscous detector guide rail fixing device through two bearings of the viscous detector guide rail fixing device; the pressure sensor probe is connected to the bottom of the pressure detector guide rail fixing device through an axial moving pair; the transverse transmission rod passes through the pressure detector through the bearing of the pressure detector guide rail fixing device The guide rail fixing device of the device; the transverse axis and longitudinal axis converter is connected to one end of the transverse transmission rod; the longitudinal transmission rod sleeve is set on the shaft hole below the transverse axis and longitudinal axis converter; the longitudinal transmission rod tip is connected to the longitudinal transmission rod sleeve through a thread; the longitudinal transmission rod tip is fixed at a preset position of the box; the loading platform is connected to the box using screws; the mechanical vision sensor is fixed at the corresponding position of the box; the flexible sensor array is fixed on the loading platform by inlaying; the door is connected to the box through a door shaft, and the control panel is installed on the box; the flexible strain sensor is pasted on the working surface below the adhesive probe cover by pasting; the flexible pressure sensor is pasted on the working surface below the pressure sensor probe by pasting; the host computer is respectively connected to the flexible pressure sensor, the mechanical vision sensor and the flexible strain sensor; the flexible sensor array is used to obtain a pressure change signal reflecting the change of pressure value. The present invention adopts a variety of sensors and detectors, including flexible sensors, pressure sensors, mechanical vision sensors, etc., which can detect the freshness of mutton with high precision. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] 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.

[0011] Figure 1 A schematic diagram of the structure of a non-destructive release detection device for mutton freshness based on multimodal data fusion and flexible sensors provided in an embodiment of the present invention;

[0012] Figure 2 A schematic diagram of a box structure provided by an embodiment of the present invention;

[0013] Figure 3 A schematic diagram of a detection method provided by an embodiment of the present invention;

[0014] Figure 4 A schematic diagram of a flow chart of a flexible sensor array locating a sample center according to an embodiment of the present invention;

[0015] Figure 5 A schematic diagram of the response process of the flexible sensor array provided in an embodiment of the present invention after receiving an external signal;

[0016] Figure 6 A schematic diagram of a flow chart for determining the number of viscosity sensor probes provided by an embodiment of the present invention;

[0017] Figure 7 A schematic diagram of a positioning and identification process of a mechanical vision device provided in an embodiment of the present invention;

[0018] Figure 8 A schematic diagram of a meat color scoring process provided by an embodiment of the present invention;

[0019] Fig. 9 A schematic diagram of a network model flow chart provided by an embodiment of the present invention;

[0020] Fig.10 A schematic diagram of the overall detection process provided by an embodiment of the present invention;

[0021] Fig.11 A schematic diagram of an information flow provided for an embodiment of the present invention.

[0022] Description of reference numerals:

[0023] 1. Flexible sensor array; 2. Horizontal transmission rod; 3. Longitudinal transmission rod sleeve; 4. Longitudinal transmission rod tip; 5. Horizontal axis and vertical axis converter; 6. Viscous detector guide rail fixture; 7. Pressure detector guide rail fixture; 8. Viscous detector integrated device; 9. Viscous probe sleeve; 10. Flexible strain sensor; 11. Pressure sensor probe; 12. Flexible pressure sensor; 13. Mechanical vision sensor; 14. Stage; 15. Box; 16. Door; 17. Control panel; 18. Host computer. DETAILED DESCRIPTION

[0024] 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.

[0025] 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.

[0026] Figure 1 A schematic diagram of the structure of a non-destructive release detection device for mutton freshness based on multimodal data fusion and flexible sensors provided in an embodiment of the present invention is shown in FIG. Figure 1 and Figure 2As shown in FIG. 1 , the device includes a flexible sensor array 1, a transverse transmission rod 2, a longitudinal transmission rod sleeve 3, a longitudinal transmission rod top 4, a transverse axis and a longitudinal axis converter 5, a viscous detector guide rail fixing device 6, a pressure detector guide rail fixing device 7, a viscous detector integrated device 8, a viscous probe sleeve 9, a flexible strain sensor 10, a pressure sensor probe 11, a flexible pressure sensor 12, a mechanical vision sensor 13, a stage 14, a box 15, a door 16, a control panel 17 and a host computer 18; the flexible sensor array 1 includes 1600 tiny sensors evenly distributed in a sensor area of ​​400 mm×400 mm. The invention relates to a sensor array, a sensor protection layer, and a signal output terminal; the micro sensor is a pressure sensor, and its function is to respond to the pressure value change above it; the sensor protection layer is made of a relatively thin flexible material; the flexible strain sensor includes a metal electrode, a flexible base film, and a signal output terminal; the manufacturing method of the flexible pressure sensor is the same as that of the flexible strain sensor, but the electrode shape and material are different; the box includes a lighting system and fixed positions of each device; the viscous probe cover is connected to the viscous detector integrated device by an axial moving pair; the viscous detector integrated device is fixed to the viscous detector fixing device by a thread The bottom of the box body is provided with a plurality of bearings, wherein the transverse transmission rod passes through the two bearings of the viscous detector fixing device and penetrates the viscous detector fixing device at the upper part of the viscous detector fixing device; the pressure sensor probe is connected to the bottom of the pressure detector fixing device through an axial moving pair; the transverse transmission rod passes through the pressure detector fixing device through the bearing of the pressure detector fixing device; the transverse axis and longitudinal axis converter is connected to one end of the transverse transmission rod; the longitudinal transmission rod sleeve is connected through the axial hole below the transverse axis and longitudinal axis converter; the top of the longitudinal transmission rod is connected to the longitudinal transmission rod sleeve through a thread, and the thread length is relatively long, and the length can be changed by twisting the thread; the top of the longitudinal transmission rod is fixed at the corresponding position of the box body; the stage is connected to the box body with screws; the mechanical vision sensor is fixed at the corresponding position of the box body; the flexible sensor array is fixed on the stage by inlaying, and its geometric center coincides with the geometric center of the stage; the door is connected to the box body through a door shaft, and the control panel is installed at the corresponding position of the box body; the flexible strain sensor is pasted on the working surface below the viscous sensor probe by pasting; the flexible pressure sensor is pasted on the working surface below the pressure sensor probe by pasting.

[0027] Specifically, the function of the metal electrode is to detect the viscosity of the meat through the objective connection between its own strain and the viscosity of the meat; the second signal output end is used to derive the signal generated by the metal electrode.

[0028] Specifically, the built-in LED light in the box 15 is turned on for lighting, and the flexible sensor array 1 is used to establish the XY coordinates of the device. When the unknown piece of meat is placed on the flexible sensor array, the piece of meat will cause pressure on the micro sensor below, and the electrical signal of the micro sensor in this area will change, thereby detecting the specific placement of the meat. After obtaining the coordinates of the unknown piece of meat, the longitudinal transmission rod sleeve 3, the longitudinal transmission rod top 4 and the transverse transmission rod 2 move the viscosity detector integrated device 7 to the geometric center of the meat piece, and the viscosity probe sleeve 6 moves downward and stops when it contacts the surface of the meat, and then moves in the opposite direction. When the viscosity probe sleeve moves upward, the flexible strain The deformation rate of the sensor increases, causing the electrical signal it transmits to change. When the strain of the flexible strain sensor reaches the maximum value and separates from the meat surface, the viscous detector fixture moves to the edge of the box, and the pressure detector fixture moves to the geometric center of the meat. The pressure sensor probe 11 drops rapidly and slows down when it contacts the meat surface. When the pressure signal transmitted by the flexible pressure sensor 12 reaches a threshold, the pressure sensor probe 11 lifts up. At this time, the mechanical vision sensor 13 detects the pit response time and grades the meat color. All signals are output to the host computer 18 for classification through the signal output end of the detection part.

[0029] Optionally, the interior of the box 15 is white or a bright color other than red, which makes the color of the meat more vivid, contrasts with the environment, and is conducive to the recognition operation of mechanical vision. There are fixing ports for the longitudinal transmission rod on both sides of the upper part of the box, which are used to fix the above-mentioned transmission device. The bottom of the box is provided with a boss design, which increases the circuit space of the flexible sensor array while reducing consumables. There are grooves at the midpoints of the four sides of the stage, which are also designed for this purpose. There is a camera probe fixing device in the middle of the right side of the box, and there are lighting device fixing devices on the left and back sides.

[0030] When the viscous sensor probe is pressed down, the surface of the flexible strain sensor can be in contact with the surface of the meat. When the probe is pulled up, the strain degree of the flexible strain sensor increases and reaches the maximum when it is separated from the meat surface. The maximum strain of the flexible strain sensor when the probe is pulled up can be used to estimate the viscosity characteristic index of the meat; the pressure sensor presses down the same size on the meat surface, and the visual sensor is used to judge the time for the pit to recover, thereby estimating the elastic characteristic index of the meat. The mechanical visual sensor is composed of an image capture device and an image processing device. The mechanical visual sensor is composed of an image capture device and a signal transmission device. Its working principle is to collect image information at regular intervals through an image capture device such as a camera. The information is in RGB three-channel mode and is transmitted to the image processing device through a signal transmission device. The image processing device is a neural network developed based on pytorch. Its working principle is: input an image, and judge its meat color characteristics through the processing of the neural network. The sensor array module is composed of a pressure flexible sensor array and a protective film. Its working principle is: take a vertex of the flexible sensor array as the origin to establish an XY coordinate system. When the meat is placed on the loading plate, the meat will give the flexible sensor array a pressure signal. The position of the meat is determined by the coordinates of the pressure signal in the XY coordinate system, and the height information of the meat is predicted by the size of the pressure signal per unit area.

[0031] The lighting device fixture is fixed with a lighting device, which consists of 6 LED lamp tubes, which are placed on the upper, middle and lower parts of the left rear side of the fixed box respectively. The working part power management module is connected to all the above working modules.

[0032] The probe working surface of the adhesive probe cover has a large radius of curvature, which is designed to imitate the curvature of a human finger and reduce the damage of the probe to the flesh. The flexible strain sensor is made of PDMS, PI, PEN or PET materials by laser direct writing or inkjet printer technology, and the electrode material is graphene or metal electrode. The sensor is connected to the sensor probe by adhesion at both ends. The pressure sensor probe is made of PDMS, PI, PEN or PET materials by laser direct writing or inkjet printer technology, and the electrode material is graphene or metal electrode.

[0033] The pressure sensor probe has a small curvature radius on the working surface of the probe, which is intended to cause a large local deformation of the meat, so as to better determine the rebound coefficient of the meat. The sensor is connected to the sensor probe by bonding at both ends.

[0034] The flexible sensor array is a 400×400 pressure sensor on the operating table. The coordinates are defined using the plane XY rectangular coordinate system. The system defaults to defining the leftmost pressure sensor facing the operating table closest to the door as the (0,0) coordinate, with the right and the inside of the operating table being the positive directions of the X and Y axes respectively. Each pressure sensor converts the information on the magnitude of the pressure it is subjected to into a signal and transmits it to the system as a data source for subsequent processing. The coordinates of the sensors with significantly increased pressure values ​​will be gathered as the spatial position range for detecting the sample meat. The system will repeatedly measure the pressure information three times, and the overlapping coordinates of each coordinate set will be taken as the position coordinates of the measured sample meat. The coordinate information is stored in the single-chip microcomputer for subsequent control of the sensor measurement. At the same time, the average pressure value on the sensor of these coordinates is taken for processing, and the height information h of the meat is predicted by the size of the pressure signal per unit area. s , which is also stored in the system.

[0035]

[0036] Among them, P s is the pressure per unit volume; A o is the unit area of ​​the pressure sensor; ρ meat The density of the meat being measured (set according to the average density of meat samples of this variety).

[0037] Preferably, the positioning of each sensor is determined by the top fixture, which is driven by electricity as the program runs, and moves along the lateral, longitudinal, and converter transmission rods, and each moving unit is consistent with the minimum unit size of the pressure sensor array. The moving control plane composed of the transmission rods constitutes a planar XY rectangular coordinate system. Preferably, when the system is initialized, the sensor fixture moves to the (0,0) coordinate just above the pressure sensor, which is the (0,0) coordinate of this coordinate system. The calibration method is that the moving pressure sensor probe has and only has a sudden change in the pressure signal with the (0,0) pressure flexibility sensor. It is stipulated that each moving unit represents a change in the coordinate value, and the coordinates are used to accurately represent the position P of the sensor probe. sensor (x k ,y k ) input is stored in the microcontroller, making it easy to run the program to control the movement of the sensor probe for measurement.

[0038] Preferably, a neural network model is used to fit the coordinate point set reflecting the position of the sample meat with a regular shape such as a rectangle or a circle. Since the selected sample meat is generally close to a regular shape, this method facilitates the selection of sampling points, and the sampling points are also representative. Then the geometric center of the fitted shape is taken as the center point P c(x0, y0), if it is a rectangle, the four points of the diagonal are taken as sampling points; if it is a circle, the four points of the diameter parallel to the X and Y axes are taken as sampling points. There are 5 sampling points in total, and the coordinates are stored in the system to prepare for the detection process.

[0039] Preferably, a meat viscosity test is performed. When the elasticity test is completed, the elasticity test probe automatically rises and stops, and the transmission device stops for 5 seconds. During this period, the operator can terminate the test, retest, change the order, etc. If the operator does not change the process, the system automatically drives the holder of the viscosity sensor probe to the top of the sampling center point of the sample meat, and the control movement process is consistent with the process in the elasticity test experiment. After that, the probe is controlled to descend to approach the sample meat, and the initial descent speed is initialized to 10-20 mm / s. Preferably, since the extended part of the viscosity test probe is long and the integrated device needs to be controlled to adjust the distance between each probe, the descent speed is slowed down and the reserved height is increased. When the predicted experimental height position is reached, the integrated device controls the rotation of the probe and the extension of the connector so that the four probes are matched to the nearest sampling sample point respectively, so that they are located directly above the sample point. Preferably, in this experiment, the present embodiment selects sample points other than the center point, so that simultaneous measurement can be achieved, which can reflect the overall viscosity level of the sample meat. After the adjustment is completed, the probe continues to descend at a certain speed, and each probe is activated at the same time.

[0040] When the probe contacts the surface of the sample meat, the probe is controlled to continue to drop 3mm, then stop for 0.5 seconds and then rise at a speed of 2-3mm / s. The information of the probe force and the rising height is recorded all the time, and stored in the system for evaluating the viscosity of the sample meat. When the probe force is recorded as 0, the probe accelerates to rise, and after rising to a certain height, the four probes are driven to restore the default position. After the recovery is completed, the connecting rod quickly rises to the initial position to prepare for the next experiment.

[0041] Preferably, the transmission device between the viscosity sensor probe and the viscosity detector integrated device is a hydraulic transmission, which has the advantages of stepless speed regulation, stability and small size.

[0042] Preferably, the flexible film material of the flexible strain sensor is PDMS film material, and the electrode material is graphene material electrode.

[0043] Preferably, the flexible film material of the flexible pressure sensor is PDMS film material, and the electrode material is metal electrode such as gold, silver, copper, etc. (determined according to needs).

[0044] Preferably, the pressure sensor probe and the pressure detector guide rail fixing device adopt a hydraulic transmission device, which has the advantages of stepless speed regulation, stability, small size and easy control.

[0045] Preferably, the image processing device adopts a neural network computing method developed based on the pytorch platform; its advantage is that it can quickly identify, analyze image features, and give classification.

[0046] Preferably, in order to realize the change of the longitudinal and lateral distances of the viscosity detector integrated device and the axial movement of the viscosity probe and the pressure probe, this embodiment adopts single-chip microcomputer control and electric drive. This embodiment first makes an overall plan, clearly defines the goal of moving and retracting the drive device, and clearly defines the requirements of the device's retractable range, speed, accuracy, etc., and selects appropriate single-chip microcomputers and related components based on these requirements, and designs hardware and software functions accordingly.

[0047] Preferably, the present embodiment selects Arduino single-chip microcomputer, which is easy to operate, and ensures that it has enough input and output pins and computing power to control the movement of the device. At the same time, a position sensor (such as an encoder or a displacement sensor) is selected to monitor the telescopic movement position of the device to obtain the required data. In order to achieve drive, the present embodiment selects a stepper motor as a driver, which can be accurately positioned at a fixed step angle and can be applied to applications requiring high-precision position control. The stepper motor also does not need a position feedback device, saving cost and complexity. Afterwards, the single-chip microcomputer, sensor and driver are connected by designing a suitable circuit, and a suitable level conversion circuit and power supply are used to ensure the normal operation of the circuit.

[0048] Preferably, in order to realize electric drive control, first, the power supply voltage and current required by the single chip microcomputer and the electronic device are connected to the power supply through an appropriate power supply interface (such as a battery, a socket or a power module).

[0049] Design circuits to drive electronic components according to the needs of the equipment. Use appropriate circuit components (such as transistors, capacitors, resistors, integrated circuits, etc.) to control current and voltage to meet the working requirements of the equipment.

[0050] Preferably, in order to achieve precise positioning, it is necessary to accurately control the motor or other peripherals. This embodiment uses components such as a microcontroller, a driver, and a sensor to achieve fine regulation and monitoring of electricity.

[0051] Preferably, in order to ensure safety, components such as voltage stabilizers, overload protection and power supply filtering are installed in the circuit, and appropriate power management circuits are designed and implemented to protect the device from damage caused by power supply problems such as overcurrent and overvoltage.

[0052] Finally, testing and debugging are carried out to ensure that the power can be connected and the power drive device can be realized, so that the device can move and retract as expected and meet the required accuracy and speed.

[0053] like Figure 3As shown, the present invention also provides a method for detecting the freshness of mutton, the method comprising:

[0054] Step 201: Use a viscosity detector to detect the viscosity of mutton.

[0055] Step 202: Use a pressure detector to detect the elasticity of the mutton.

[0056] Step 203: Use a machine vision sensor to detect the color of the mutton.

[0057] Step 204: Utilize the signal output ends of the three sensors to send the detection signals to a host computer for processing.

[0058] Step 205: The freshness grade of the mutton is displayed on the display on the control panel according to the calculation result. The freshness grade and the score in the host computer can be displayed simultaneously.

[0059] In practical applications, the center coordinates of the sample are determined by a flexible sensor array, and the center point iterative search method is used to approximately obtain the sample center required for the detection experiment. Figure 4 As shown, the specific steps are as follows:

[0060] Step 1: Initialization: Since there is only one center point to be found, a data point is randomly selected as the center point in the initial state.

[0061] Step 2: Assign data points: For each data point, calculate its distance from the center point and assign it to the set represented by the center point.

[0062] Step 3: Update the center point: Calculate the average of all data points in the set and use the average as the new center point.

[0063] Step 4: Repeat steps 2 and 3 until the center point no longer changes or the predetermined number of iterations is reached.

[0064] Step 5: Output result: Finally, the center point P of the set is obtained c (x0,y0).

[0065] Step 6: Based on the obtained center point, other sampling points can be obtained according to the following formula:

[0066]

[0067] x max / y max : The maximum value of the horizontal / vertical coordinates of the sample meat coordinate points;

[0068] x min / y min : The minimum value of the horizontal / vertical coordinates of the sample meat coordinate points;

[0069] The coordinate value combination can get four sampling points. There are 5 sampling points in total, and the coordinates are stored in the system to prepare for the detection process. After receiving the operator's instruction to conduct the meat elasticity detection experiment, the transmission device starts working. First, determine the probe movement trajectory:

[0070]

[0071] The holder of the electrically driven elastic detection probe moves on the X-axis crossbar to the center point of the sample meat, and the moving distance and direction can be determined by the x-coordinate difference between the holder and the center point. Similarly, the holder of the electrically driven elastic detection probe moves on the Y-axis crossbar to the center point of the sample meat, and the moving distance and direction can be determined by the y-coordinate difference between the holder and the center point. The X-axis crossbar where the holder of the electrically driven elastic detection probe is located moves on the Y-axis longitudinal bar to make the y-coordinates of the probe and the center point of the sample meat consistent, and the moving distance and direction can be determined by the y-coordinate difference between the holder and the center point. Similarly, the holder of the electrically driven elastic detection probe moves on the X-axis crossbar to the center point of the sample meat, and the moving distance and direction can be determined by the x-coordinate difference between the holder and the center point. Then the probe is controlled to descend to approach the detection sample meat, and the initial descent speed can be initialized to 10-20 mm / s. When it reaches the predicted experimental height position, the speed slows down to 3-5 mm / s, and the pressure sensor on the probe is activated to start detecting and transmitting pressure information, and the system starts recording pressure data and the distance the probe descends. Predicted experimental height h p It can be calculated according to the following formula:

[0072] h p =h s +h b

[0073] Among them, h s is the height of the meat at the sampling point, h b is the fault tolerance height, initialized to 10mm, h b is the maximum possible error value of the meat height. Since the movement is based on the top rod in actual operation, it is only necessary to control the probe to move downward by a distance h. m That's it.

[0074] h m =H-(h p +h o )

[0075] Where H is the distance between the top rod and the top surface of the operating table, h o To restore altitude.

[0076] After the probe contacts the sampling point of the meat, when the pressure value suddenly changes, the probe continues to descend for 2 seconds to form a more obvious pit on the meat surface. If the pressure remains at a similar level after the sudden change, it is effective. The probe begins to rise, and the descending height is recorded. At the same time, the mechanical vision sensor keeps observing the sampling point and records the time t required for the pit formed by the elastic probe to return to its initial state. r , stored in the system as raw data for subsequent numerical calculations of elasticity. After rising to the predicted experimental height of this test, the control probe moves to the top of the next sampling point. The detection order can be specified separately. The default detection order is counterclockwise from the point above the center point (left). The experimental data collection process is consistent. The collected data is stored in the system.

[0077] After each measurement is completed, the sensor recovers and rises to its original position. This way, the initial height of each sensor during measurement will not be affected by the different properties of the meat, such as thickness, viscosity, elasticity, etc., which facilitates the next measurement.

[0078] In actual application, when the detector reaches the target area, the microcontroller controls the detector probe to descend, and the sensor on the probe outputs a signal. When the signal reaches the predetermined standard, the detector probe rises. The specific steps are as follows: Figure 5 shown.

[0079] In practical applications, the shape and area of ​​the meat piece to be tested for freshness are judged by the sensor array, and the flexible strain sensor determines the number of flexible strain sensor probes to be used according to the signal given by the sensor array, and changes the distance between the four probes and the center of the flexible strain sensor by telescoping to ensure that the viscosity measurement is true and effective.

[0080] Specifically, according to the circumscribed circle area when the four probes of the flexible strain sensor can be retracted at the minimum, the maximum inscribed circle area in the shape of the meat block to be measured obtained by the sensor array is divided into four ranges, from small to large, formed by three thresholds: a first inscribed circle area threshold, a second inscribed circle area threshold, and a third inscribed circle area threshold; the sensor array obtains the shape of the meat block and its size, and judges the size relationship between the maximum inscribed circle area in the shape and the three inscribed circle area thresholds to obtain a judgment result; if the judgment result obtained indicates that the inscribed circle area of ​​the meat block to be measured is smaller than the first inscribed circle area threshold, only one of the four probes of the above-mentioned flexible strain sensor is used to perform four viscosity measurements at different positions on the meat block; if the judgment result obtained indicates that the inscribed circle area of ​​the meat block to be measured is larger than the first inscribed circle area threshold and smaller than the second inscribed circle area threshold Circle area threshold, use two of the four probes of the flexible strain sensor to perform two viscosity measurements at different positions of the meat block, and the distance between the four probes of the flexible stress sensor is the average of the maximum elongation and the minimum elongation; if the judgment result indicates that the inscribed circle area of ​​the tested meat block is greater than the second inscribed circle area threshold and less than the third inscribed circle area threshold, use three of the four probes of the flexible strain sensor to perform two viscosity measurements at different positions of the meat block, and the distance between the four probes of the flexible stress sensor is the average of the maximum elongation and the minimum elongation; if the judgment result indicates that the inscribed circle area of ​​the tested meat block is greater than the third inscribed circle area threshold, the relative distances of the four probes can be changed according to the formula, and the meat block is measured for one viscosity measurement to ensure the effectiveness of the four-point sampling viscosity measurement.

[0081] The flexible strain sensor changes the relative positions of the four probes according to the obtained meat information, determines the selected points in the four-point sampling method, and moves the flexible strain sensor to the position to be measured through the mechanical transmission device. The formula for the relative center distance of the four probes of the strain sensor is:

[0082]

[0083] First inscribed circle area threshold:

[0084] Second inscribed circle area threshold:

[0085] The third inscribed circle area threshold: S3 = πl min 2 .

[0086] Among them, l min is the minimum relative center of the four probes of the flexible strain sensor, L is the perimeter of the meat block to be tested, δ is the maximum inscribed circle radius of the meat block to be tested, S is the area of ​​the meat block to be tested, S1 is the first area threshold, S2 is the second area threshold, S3 is the third area threshold, S内 is the maximum inscribed circle area of ​​the meat to be tested. Figure 6 As shown in FIG. 1 , the stretchability is minimal when the stretch amount is 0 (no stretch).

[0087] like Figure 7 As shown in the figure, in actual application, after the piece of meat to be inspected is placed on the stage, the mechanical vision sensor starts working. The system controls the sensor to rotate regularly and shoot the stage. First, the image of the captured picture is grayed out. The RGB channel values ​​of each pixel of the color image are linearly combined according to certain weights to obtain the corresponding grayscale value. Since the human eye is most sensitive to green and least sensitive to blue, in order to be closer to the observation effect of the human eye, the weighting of the three RGB components is different, and the calculation is shown as follows:

[0088] V gray (x, y) = 0.299 V red (x, y)+0.578·V green (x, y)+0.114·V blue (x, y)

[0089] After grayscale processing, the gradient of each pixel is calculated. This embodiment uses an edge detection operator and uses the following convolution kernels for image A in the horizontal and vertical directions:

[0090]

[0091] The gradients of the image in the horizontal and vertical directions are calculated using L x , L y express.

[0092] Then use the intermediate matrix Q

[0093]

[0094] Where p(x,y) is the window function used for weighted summation of pixels, and q(x,y) is the introduced correction function.

[0095] The next step is to calculate the edge corner response function A p :

[0096] A p =detQ-k(traceQ) 2

[0097] detQ=β1β2

[0098] traceQ=β1+β2

[0099] After calculating the edge corner response values ​​of all pixels, non-maximum suppression is performed to retain the local maximum value to eliminate redundant corner points. Then a threshold is set according to the corner response value, and only pixels greater than the threshold are retained. Pixels that pass the threshold screening are marked as corner points. And the coordinates of the corner points in the image are obtained.

[0100] In the last step of data processing, the coordinates of all detected corner points are calculated and averaged to estimate the center point of the target object. The center coordinates of the image taken by the camera are compared to calculate the offset of the target object in the image.

[0101] Next, according to the offset of the target object in the image, it is mapped to the signal that controls the direction and angle of the camera. The direction and angle of the camera can be controlled to keep the target object in the center, that is, the average coordinates coincide with the center of the image. After the angle and height are adjusted, adjust the focus and the proportion of the meat in the image until it is around 70%.

[0102] After obtaining an accurate image of the detected meat, this embodiment will perform meat color detection. First, this embodiment converts the pixel value in the RGB color space into the HSV (hue, saturation, brightness) color space.

[0103] First, the pixel value (R, G, B) in the RGB color space is normalized to the range [0, 1] to obtain the new value (r, g, b).

[0104]

[0105] The conversion calculation obtains the (h, s, v) value of each pixel in the HSV space. The conversion calculation formula is:

[0106] v=max{r,g,b}

[0107]

[0108]

[0109] The next step is to determine the freshness of the test cuts of meat by comparing the color difference between the test cuts and fresh cuts of meat.

[0110] First, it is necessary to determine the representative HSV value of each piece of meat to represent the overall color of the piece of meat. This embodiment uses the center point iteration to find a color representative value to determine the color of the piece of meat. The steps are as follows:

[0111] 1) Data processing: Calculate the average H, S, and V values ​​of the color data points in the meat block respectively, and use the (h, s, v) point as the initial representative value.

[0112] 2) For each data point, calculate its distance from the initial color representative value, and then assign the data point to the cluster with the closest color representative value.

[0113] The distance calculation formula is used here:

[0114]

[0115] (h i ,s i ,v i ) is the HSV value of each point, (h o ,s o ,v o ) is the HSV value of the center point

[0116] 3) Update the color representative value: For each cluster, calculate the average value of all data points in the cluster and use this average value as the new color representative value.

[0117] 4) Repeat iteration: Repeat steps 2 and 3 until the color representation value does not change on the order of 10^-3.

[0118] 5) Obtain the (h, s, v) color value representing the overall color of the meat.

[0119] Secondly, the color of standard fresh meat pieces is worth obtaining.

[0120] First, 50 to 100 pieces of freshly slaughtered meat that are confirmed by professional inspectors to be of excellent freshness are obtained, and the image of each piece of meat is subjected to the above-mentioned clustering algorithm to determine a color representative value of the color of the piece of meat. Then, these representative values ​​are used as a data set, and the above-mentioned iterative algorithm is performed again. The color representative value finally obtained is the color (h, s, v) value of the standard fresh meat block recognized in this embodiment.

[0121] The last step is to establish a colorimetric system.

[0122] A batch of freshly slaughtered meat is obtained and stored under certain conditions. Each piece of meat is tested every certain period of time (e.g., 12 hours) to obtain the (h, s, v) value, and the distance data from the color (h, s, v) value of the standard fresh meat is calculated until the meat is completely corrupted. A distance value is obtained for each experiment by taking the average value, which is divided into 6 levels or 6 intervals from excellent freshness to complete corruption, with corresponding scores of 5, 4, 3, 2, 1, and 0.

[0123] Due to differences in storage conditions and the conditions of the meat itself, the colorimetric system needs to be re-established in different specific situations. Figure 8 shown.

[0124] During the experimental test, the score M is calculated by calculating the distance value:

[0125]

[0126] where d i : is the distance between the detected meat block and the standard fresh meat block, d0: is the distance between the 0-grade meat block and the standard fresh meat block, d5: is the distance between the 5-grade meat block and the standard fresh meat block.

[0127] If the color is detected to be white, that is, there are more similar parts of the RGB three values, the weight of the color score in calculating the total score is reduced. The calculation method of the weight W of the color score is:

[0128] W = 0.83 (1-w white )

[0129] where w red This is to detect the proportion of the white part in the color to the overall color. For example, this step is in parallel with the above step, which is responsible for locating the meat block, and this step is the color detection of the meat block.

[0130] In practical applications, the obtained flesh color parameters are the image information of the RGB three channels, the obtained elastic parameters are the time information of the pit recovery, and the obtained viscosity parameters are the electrical signal information transmitted by the flexible strain sensor on the viscosity detector probe. The processing of the flesh color parameters has been explained in detail above. Figure 7 As shown, the elasticity parameter is processed by comparing with the indicators of the standard sample, from excellent freshness to complete corruption, divided into 6 levels or 6 intervals, with corresponding scores of 5, 4, 3, 2, 1, and 0. The viscosity parameter is first stabilized by a voltage follower device to stabilize the output voltage of the flexible strain sensor, and then an operational amplifier circuit is added to amplify its characteristics, and compared with the annual parameters of the standard sample, from excellent freshness to complete corruption, divided into 6 levels or 6 intervals, with corresponding scores of 5, 4, 3, 2, 1, and 0. Using multimodal data fusion, the elasticity, viscosity, and meat color scores are input into the config module. The config module mainly realizes the combination of multimodal data through code so that it can be input into the subsequent network together. The config backend network is the MyModel model, and its specific network structure is as follows Fig. 9As shown, the results are produced through convolution, pooling, linear and nonlinear operations. Specifically, first, the meat color information is convolved and pooled to obtain its features, and the features are input into the detect module to analyze its color. If the detect module detects it as white, it is fed back to the liner layer to change liner (83) to liner (20). This operation can reduce the weight of the model occupied by meat color when there is more fat on the surface of mutton. Then, the feature values ​​of viscosity, elasticity, and meat color are fused in the same linear function through the cat instruction. After two convolution operations, a maximum pooling operation is performed to fuse the data before convolution pooling with the data after convolution pooling in a certain proportion to complete the features. After repeating this step twice, liner (6) processing is performed to obtain a list of length six. The 0-5 in the list represent the probability of predicting the freshness of the meat to be at that level. By comparing with the actual freshness of the meat, the data is fed back to the convolution kernel of the convolution step to modify the data in the convolution kernel. After at least 300 iterations, the prediction result becomes stable.

[0131] Specifically, the Conv2d module and the maxpool layer are used to extract features. The middle liner layer is used for linearization, so that the image information and non-image information can be converted into tensors of the same dimension for better fusion. The Totensor layer converts the input information into tensor form. The cat layer combines the linearized tensors with the last dimension. The last liner layer is used to derive the probability of the classification label. Its output is a list of length 6, and the elements in the list represent the probability values ​​of each label.

[0132] Convolution formula:

[0133]

[0134] (where ε is the brightness value of the flesh color information at that matrix point, and τ is the value of the convolution kernel at that matrix point)

[0135] In practical applications, the detection process diagram is as follows Fig.10 shown.

[0136] In practical applications, information flow diagrams such as Fig.11As shown, input 1 is the input signal of the power on button. In the power-on link, the signal starts the power supply and outputs to the LED, the flexible pressure sensor array and the visual sensor to make them in working state. Output 1 is the light signal of the LED lighting. After receiving the power-on signal and input 2 (the pressure signal generated by the meat being put in), the flexible sensor array transmits the signal to the threshold analysis system and the sampling point positioning system respectively. The sampling point positioning system transmits the information to the viscosity and elasticity detectors. The threshold analysis system transmits the signal to the viscosity detector after processing. After the input of these two information, the viscosity and elasticity detectors collect the viscosity and elasticity information of the meat respectively as signals and output them to the config module. The config module fuses and outputs them, fuses them with the information collected by the visual sensor, and inputs them into the MyModel model for analysis and processing. The model outputs a list of 6 in length, which respectively represent the probability information of the freshness predicted as 0, 1, 2, 3, 4, and 5 levels. The information is processed to obtain the freshness level as output 2. At this point, the freshness detection process is completed, and the user obtains the freshness level of the meat.

[0137] 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.

[0138] 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 bionic detection device for mutton freshness, characterized in that: include: Flexible sensor array, transverse transmission rod, longitudinal transmission rod sleeve, longitudinal transmission rod top, transverse axis and longitudinal axis converter, viscous detector guide rail fixture, pressure detector guide rail fixture, viscous detector integrated device, viscous probe sleeve, flexible strain sensor, pressure sensor probe, flexible pressure sensor, mechanical vision sensor, stage, box, door, control panel and host computer; The viscous probe cover is connected to the viscous detector integrated device through an axial moving pair; the viscous detector integrated device is fixed below the viscous detector guide rail fixing device through threads; the transverse transmission rod passes through the entire viscous detector guide rail fixing device through two bearings of the viscous detector guide rail fixing device; The pressure sensor probe is connected to the lower part of the pressure detector guide rail fixing device through an axial moving pair; the transverse transmission rod passes through the pressure detector guide rail fixing device through the bearing of the pressure detector guide rail fixing device; the transverse axis longitudinal axis converter is connected to one end of the transverse transmission rod; the longitudinal transmission rod sleeve is sleeved on the shaft hole below the transverse axis longitudinal axis converter; the longitudinal transmission rod top is connected to the longitudinal transmission rod sleeve through a thread; the longitudinal transmission rod top is fixed at a preset position of the box; the stage is connected to the box with screws; the mechanical vision sensor is fixed to the corresponding position of the box; the flexible sensor array is fixed to the stage by inlaying; the door is connected to the box through a door shaft, and the control panel is installed on the box; the flexible strain sensor is pasted on the working surface below the adhesive probe cover by pasting; the flexible pressure sensor is pasted on the working surface below the pressure sensor probe by pasting; the host computer is respectively connected to the flexible pressure sensor, the mechanical vision sensor and the flexible strain sensor; the flexible sensor array is used to obtain a pressure change signal reflecting the change of pressure value.

2. The non-destructive bionic detection device for mutton freshness according to claim 1 is characterized in that: The geometric center of the flexible sensor array is the same as the geometric center of the stage.

3. The non-destructive bionic detection device for mutton freshness according to claim 1 is characterized in that: The flexible sensor array includes multiple microsensors, a sensor protection layer and a first signal output end; the microsensor is a pressure sensor; the sensor protection layer is made of a flexible material; the sensor protection layer is arranged on the outer shell of the microsensor; the microsensor is used to obtain the pressure change signal; the first signal output end is respectively connected to the microsensor and the host computer.

4. The nondestructive bionic detection device for mutton freshness according to claim 1 is characterized in that: The flexible strain sensor comprises a metal electrode, a flexible base film and a second signal output terminal; The metal electrode is attached to the top of the flexible base film by 3D printing technology, and the second signal output end is welded to both sides of the metal electrode by welding; the second signal output end is connected to the host computer by a wire.

5. The non-destructive bionic detection device for mutton freshness according to claim 1 is characterized in that: The box body is provided with an LED lamp built therein; the LED lamp is used to illuminate the inner space of the box body.

6. A non-destructive bionic detection method for mutton freshness, characterized in that: The non-destructive bionic detection device for mutton freshness applied to any one of claims 1 to 5, the method comprising: The target detection meat block is placed on the flexible sensor array to generate a pressure change signal; Based on the host computer, the coordinates of the target detection meat block are determined according to the pressure change signal, and the geometric center and shape area information of the target detection meat block are determined according to the coordinates of the target detection meat block; Control the transverse transmission rod, the longitudinal transmission rod sleeve and the longitudinal transmission rod tip to move, so as to drive the viscous probe integrated device to move to the geometric center of the target detection meat block, and control the viscous probe sleeve to move downward, so as to stop when the flexible strain sensor contacts the surface of the target detection meat block, and control the viscous probe sleeve to move upward after stopping; When the sticky probe cover moves upward, the deformation rate of the flexible strain sensor increases, so that the electrical signal transmitted by the flexible strain sensor changes. When the strain of the flexible strain sensor reaches the maximum value and leaves the meat surface, the sticky detector guide rail fixing device is controlled to move to the edge of the box; Control the pressure detector fixing device to move to the geometric center of the target detection meat block, and control the pressure sensor probe to descend, and reduce the speed when the pressure sensor probe contacts the surface of the target detection meat block, and control the pressure sensor probe to lift up when the pressure signal transmitted by the flexible pressure sensor reaches a preset threshold; In the process of the pressure sensor probe being lifted upward, the mechanical vision sensor and the host computer are used to detect the response time of the pit of the target detection meat block and the color of the target detection meat block; Performing viscosity index grading according to the signal of the flexible strain sensor to obtain a viscosity grading result, performing color grading according to the color of the target detected meat block to obtain a color grading result, and performing elasticity index grading according to the recovery time of the pit of the target detected meat block to obtain an elasticity grading result; Based on a multimodal data fusion method, the freshness grade of the target detected meat block is determined according to the viscosity grading result, the color grading result and the elasticity grading result.

7. The nondestructive bionic detection method for mutton freshness according to claim 6, characterized in that: Based on the host computer, the coordinates of the target detected meat block are determined according to the pressure change signal, including: Randomly select a data point from the coordinate point set of the target detected meat block as the center point; For each data point, select a data point as the center point, calculate the distances from other data points to the center point, and assign the data point to the set represented by the nearest center point based on the calculated distances; Calculate the average of the horizontal and vertical coordinates of all data points in the set represented by each center point, and use the average as the new center point; Repeat the steps of "for each data point, select a data point as the center point, calculate the distances from other data points to the center point, and assign the data point to the set represented by the nearest center point according to the calculated distance" and "calculate the average of the horizontal and vertical coordinates of all data points in the set represented by each center point, and use the average as the new center point". After multiple iterations, when the four center points converge to the same coordinate or the specified number of iterations is reached, the iteration stops, and the center point of each set converges to the true center point of the data point set to obtain the final center point; The coordinates of other sampling points are calculated based on the final center point; the calculation formula for the coordinates of the other sampling points is: Among them, x max and max are the maximum values ​​of the horizontal and vertical coordinates of the target detection meat block coordinate points, x min and min are the minimum values ​​of the horizontal and vertical coordinates of the target detection meat block, respectively. i is the horizontal coordinate value of other sampling points, y i is the ordinate value of other sampling points, x0 is the abscissa value of the final center point, and y0 is the ordinate value of the final center point; The coordinates of each other sampling point and the final center point are stored to determine the coordinates of the target detection meat piece.

8. The nondestructive bionic detection method for mutton freshness according to claim 6, characterized in that: Before controlling the downward movement of the adhesive probe cover, it also includes: Determine the maximum inscribed circle area of ​​the target meat block according to the shape area information; Divide the maximum inscribed circle area according to the circumscribed circle areas when the probes of the flexible strain sensor can be stretched to the minimum, and obtain a first inscribed circle area threshold, a second inscribed circle area threshold, and a third inscribed circle area threshold with values ​​from small to large; Determine the relationship between the maximum inscribed circle area and the first inscribed circle area threshold, the second inscribed circle area threshold, and the third inscribed circle area threshold to obtain a determination result; If the judgment result indicates that the maximum inscribed circle area is less than the first inscribed circle area threshold, one of the several probes of the flexible strain sensor is used to perform four viscosity measurements on the target detection piece of meat at different positions; if the judgment result indicates that the maximum inscribed circle area is greater than the first inscribed circle area threshold and less than the second inscribed circle area threshold, two of the several probes of the flexible strain sensor are used to perform two viscosity measurements on the target detection piece of meat at different positions; if the judgment result indicates that the maximum inscribed circle area is greater than the second inscribed circle area threshold and less than the third inscribed circle area threshold, three of the several probes of the flexible strain sensor are used to perform two viscosity measurements on the target detection piece of meat at different positions; if the judgment result indicates that the maximum inscribed circle area is greater than the third inscribed circle area threshold, one viscosity measurement is performed on the target detection piece of meat using several probes of the flexible strain sensor; The formula for the relative center distance of several probes of the flexible strain sensor is: The formula for the first inscribed circle area threshold is: The formula for the second inscribed circle area threshold is: The formula for the third inscribed circle area threshold is: S3 = πl min 2 ; The formula for determining the number of probes is: n = δ 2 (2l 2 min ); Where l is the distance from each probe to the center, l min is the distance of the probe of the flexible strain sensor relative to the center when the stretchability is the minimum, L is the circumference of the target detection meat block, δ is the maximum inscribed circle radius of the target detection meat block, S is the area of ​​the target detection meat block, S1 is the first inscribed circle area threshold, S2 is the second inscribed circle area threshold, S3 is the third inscribed circle area threshold, S 内 is the maximum inscribed circle area of ​​the target meat block, and n is the number of the flexible strain sensor probes.

9. The nondestructive bionic detection method for mutton freshness according to claim 6, characterized in that: The method for detecting the color of the target meat block comprises: The RGB channel values ​​of each pixel of the color image captured by the machine vision sensor are linearly combined according to the preset weights to obtain the corresponding grayscale value; the grayscale value is calculated as follows: V gray (x, y) = 0.299 V red (x, y)+0.578·V green (x, y)+0.114·V blue (x, y); finally, the grayscale processed image is obtained; where V red is the pixel red channel value, V green is the value of the pixel green channel, V blue is the value of the pixel green channel, V gray is the pixel gray value; Using the edge detection operator, the grayscale image is calculated using the convolution kernel in the horizontal and vertical directions respectively to obtain the gradient L of the grayscale image in the horizontal direction. x and the vertical gradient L y ; The formula for the convolution kernel in the horizontal direction is: The formula for the convolution kernel in the vertical direction is Among them, I x is the convolution kernel in the horizontal direction, I y is the convolution kernel in the vertical direction, and A is the image after grayscale processing; According to the horizontal gradient L x and the vertical gradient L y Calculate the intermediate matrix Q; the calculation formula of the intermediate matrix Q is: Among them, p(x,y) is the window function used for weighted summation of pixels, and q(x,y) is the introduced correction function; Determine the edge corner response function A based on the intermediate matrix Q p ; The edge corner response function A p The calculation formula is: Where det is the determinant of the matrix Q, and k is the trace of the matrix Q; According to the edge corner response function A p Calculate the edge corner point response values ​​of all pixels, perform non-maximum suppression, retain the local maximum value, so as to eliminate redundant corner points, and set a screening threshold value according to the edge corner point response value to retain the pixels greater than the screening threshold value, and mark the pixels that pass the threshold screening as corner points and obtain the coordinates of the corner points in the image; The coordinates of all detected corner points are averaged to estimate the center point of the target object, and compared with the center coordinates of the color image taken by the machine vision sensor to calculate the offset of the target detected meat block in the image; x c is the X-axis coordinate of the center of the target meat block, y c is the Y-axis coordinate of the target detection meat block center, ε is the deviation distance between the visual center and the target detection meat block center, and n is the number of corner points; Convert the data type characteristics of the offset into signal characteristics, and control the direction and angle of the machine vision sensor according to the converted signal characteristics to keep the target detected meat piece at the center position; After the angle and height of the machine vision sensor are adjusted, the focus is adjusted, and the proportion of the target meat block in the color image is adjusted. After the ratio is adjusted to the preset ratio, the focus is stopped to obtain an accurate image; Converting pixel values ​​in the RGB color space of the accurate image into the HSV color space; The pixel values ​​in the RGB color space of the accurate image are normalized to obtain new pixel values; the normalization formula is: Among them, R, G and B are the pixel values ​​of the points corresponding to the RGB channels in the three channels of the color image, r, g and b are the normalized values ​​of R, G and B respectively; Performing a conversion calculation in the HSV space according to the new pixel value to obtain a pixel value in the HSV space; the conversion calculation formula is: v = max{r, g, b}; h, s and v correspond to the coordinate values ​​in the HSV space respectively; Perform iterative calculation of color values; the iterative calculation includes: Calculate the H, S, and V values ​​of the pixel points in the target meat block respectively, and use the (h, s, v) point as the coordinate of the point in the HSV space; randomly select four coordinates as the color representative values ​​in the initial state; For each pixel, calculate the distance between it and the color representative value, and assign the data point to the cluster with the closest color representative value according to the distance calculation formula; the distance calculation formula is: (h i ,s i ,v i ) is the HSV value of each point, (h o ,s o ,v o ) is the value of the color representative value HSV; For each cluster, calculate the average value of all data points in the cluster and use the average value as the new color representative value; Repeat steps "For each pixel, calculate the distance between it and the color data point, and assign the data point to the cluster with the closest color representative value according to the distance calculation formula" and "For each cluster, calculate the average value of all data points in the cluster, and use the average value as the new color representative value". As the number of iterations increases, the four color representative values ​​gradually converge to one point. When the color representative value does not change at the order of 10^-3, stop and obtain the (h, s, v) color value representing the overall color of the target detected meat block; Obtain 50 to 100 pieces of meat that have been slaughtered and have reached preset freshness requirements, and perform the iterative calculation on the image of each piece of meat that has reached the preset requirements to determine a color representative value of the color of the piece of meat; then use the color representative value as a data set, perform the iterative calculation again, and finally determine the obtained color representative value as the color (h, s, v) value of the standard fresh meat block; Obtain freshly slaughtered meat pieces and store them under preset conditions. Perform an experiment on each freshly slaughtered meat piece at preset intervals to obtain the (h, s, v) value, and calculate the distance data from the color (h, s, v) value of the standard fresh meat piece until the meat piece is completely corrupted. A distance value was obtained for each experiment by taking the average value, which was divided into 6 levels or 6 intervals from excellent freshness to complete corruption, with corresponding scores of 5, 4, 3, 2, 1, and 0 points; The method for calculating the score M after calculating the distance value is: where d i : is the distance between the tested meat block and the standard fresh meat block, d0: is the distance between the 0-graded meat block and the standard fresh meat block, d5: is the distance between the 5-graded meat block and the standard fresh meat block; If the color is detected to be white, that is, there are more similar parts of the RGB three values, the weight of the color score in calculating the total score is reduced. The calculation method of the weight W of the color score is: W=0.83 (1-w white ) where w red It is the proportion of the white part in the detected color to the overall color situation.

10. The nondestructive bionic detection method for mutton freshness according to claim 6, characterized in that: Based on the multimodal data fusion method, the freshness grade of the target detected meat block is determined according to the viscosity grading result, the color grading result and the elasticity grading result, including: Inputting the viscosity grading result, the color grading result and the elasticity grading result into a config module to obtain fusion information; The fusion information is input into the MyModel model for processing to obtain the freshness level of the target detected meat piece.

Images