A device and method for online detection of green tea appearance quality

Abstract

The present invention discloses an online detection device and method for the appearance quality of green tea, wherein the detection device comprises a spreading device arranged on a detection platform for spreading green tea on the detection platform, an image acquisition device for acquiring images of the spread green tea, and a driving device for driving the spreading device and the image acquisition device to move along the length direction of the detection platform; a withering trough is arranged on the detection platform, and a weighing module is arranged below the withering trough; the spreading device comprises a storage mechanism for holding the green tea to be detected and a green tea conveying mechanism arranged in the storage mechanism for quantitatively conveying the green tea to the withering trough; the image acquisition device comprises a box, a camera module arranged in the box, and an infrared detection module. The online detection device for the appearance quality of green tea of ​​the present invention can automatically spread the green tea to be detected on the detection platform, and has a higher degree of automation.

Description

Technical Field

[0001] The invention relates to a tea leaf detection device, and in particular to an online detection device and method for the appearance quality of green tea leaves. Background Art

[0002] Tea contains tea polyphenols, tea pigments, tea polysaccharides and other ingredients that are beneficial to human health. Therefore, tea is a popular natural health drink in my country. The growth of tea is affected by many internal and external factors. To ensure the quality of tea and improve its economic effect, scientific and efficient production technology is of great significance to improving the efficiency of tea production.

[0003] The quality inspection of green tea is mainly carried out by internal quality and external quality. However, at present, most of the green tea quality inspection is to determine the contents of green tea by chemical methods, and judge the quality of green tea according to its contents. This method of inspecting contents is time-consuming, consumable, and not conducive to large-scale operation. According to enterprise standards, the external quality of tea is mainly inspected by the size of water content and the ratio of buds and leaves. Different bud and leaf ratios correspond to the uniformity and tenderness of the raw materials, and many famous teas require 1 bud and 1-2 leaves. Therefore, different bud and leaf ratios also determine the quality grade of the finished tea produced; water content also plays a vital role in the quality of tea. In actual production, tea is accompanied by the loss of water, the loss of turgor pressure of withered leaf cells, the softening of leaf quality, the reduction of leaf area, and the loss of selective permeability of cell membranes. At the same time, a series of changes in the internal chemical composition also occur, which have an important impact on the material basis of tea and the formation of subsequent processing quality. However, the method of using external quality to inspect the quality of green tea in the production process is mainly through manual operation, which is not suitable for large-scale detection and has large errors.

[0004] In recent years, thanks to the rapid development of computers, machine vision technology has also begun to be widely used in various fields such as military, medicine, vehicles and agriculture. In response to these problems, the use of machine vision technology can achieve non-destructive testing, rapid analysis and real-time analysis, and can also make up for the shortcomings of traditional quality testing. Machine vision mainly judges by the appearance of tea leaves, and analyzes by extracting texture and color features. In the sensory testing of tea leaves, the features that characterize the grade of tea leaves can also play a role in quantitative description. In recent years, deep learning technology has improved this traditional feature extraction technology, extracting the internal features of the image through convolution, and maintaining the spatial information of the image, with wider applicability and higher accuracy.

[0005] However, when testing the green tea leaves, the green tea leaves need to be spread out on the testing platform. However, the existing spreading method mostly involves manually spreading the green tea leaves to be tested on the testing platform, and the degree of automation is not high. Summary of the invention

[0006] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide an online detection device for the appearance quality of green tea leaves. The online detection device for the appearance quality of green tea leaves can automatically spread the green tea leaves to be detected on a detection platform, and has a higher degree of automation.

[0007] The second object of the present invention is to provide an online detection method for the appearance quality of green tea.

[0008] The technical solution of the present invention to solve the above technical problems is:

[0009] An online inspection device for the appearance quality of green tea leaves comprises a spreading device arranged on a testing platform for spreading the green tea leaves to be tested on the testing platform, an image acquisition device for acquiring images of the green tea leaves spread on the testing platform, and a driving device for driving the spreading device and the image acquisition device to move along the length direction of the testing platform, wherein:

[0010] The detection platform is provided with a withering trough, and a weighing module is provided below the withering trough, and the weighing module is used to weigh the green tea leaves on the withering trough;

[0011] The paving device comprises a storage mechanism for holding the green tea to be tested and a green tea conveying mechanism arranged in the storage mechanism for quantitatively conveying the green tea to the withering tank on the testing platform;

[0012] The image acquisition device comprises a box, a camera module arranged in the box, and an infrared detection module.

[0013] Preferably, the storage mechanism consists of a feeding part, a conveying part and a flattening part, wherein the feeding part is funnel-shaped, and a conveying channel is arranged in the conveying part, the upper end of the conveying channel is connected with the discharge port at the lower end of the feeding part, and the lower end is connected with one end of the flattening part; the other end of the flattening part extends downwardly at a horizontal angle.

[0014] Preferably, air holes are provided on both sides of the conveying channel of the vertical portion, and the air holes are multiple and arranged in an array; the multiple air holes are connected to the air pump through an air pipe.

[0015] Preferably, a sensor is provided at the bottom of the flat portion.

[0016] Preferably, the green tea conveying mechanism includes a conveying roller and a conveying motor for driving the conveying roller to rotate, wherein the conveying roller includes a rotating shaft and a plurality of groups of partitions arranged on the rotating shaft, wherein the rotating shaft is arranged in the conveying part and extends along the width direction of the conveying part; one end of the rotating shaft is rotatably connected to the conveying part, and the other end passes through the interior of the conveying part; the conveying part is provided with a rotating bearing at a position in contact with the rotating shaft; a plurality of groups of partitions are radially arranged on the rotating shaft, and all extend along the axial direction of the rotating shaft; the angles between two adjacent groups of partitions are equal; the conveying motor is installed on the outside of the conveying part and connected to the rotating shaft.

[0017] Preferably, the paving device further comprises a lifting drive mechanism, which is disposed on the box body and is used to drive the storage mechanism to perform lifting movements; the lifting drive mechanism adopts a driving method combining a motor and a screw transmission mechanism, or adopts an electric push rod.

[0018] Preferably, the driving device includes a mobile frame and a mobile driving mechanism for driving the mobile frame to move, wherein two groups of walking wheels are arranged on the mobile frame, and the two groups of walking wheels are respectively arranged on the front and rear sides of the mobile frame, each group of walking wheels has two walking wheels, and the two walking wheels are respectively arranged on the left and right sides of the mobile frame; wherein the walking wheels located on the front and rear sides are connected by a synchronous belt transmission mechanism, and the walking wheels located on the left and right sides are connected by a connecting shaft; the mobile driving mechanism includes a mobile motor, and the mobile motor is installed on the mobile frame, and the output shaft of the mobile motor is connected to one of the connecting shafts through a synchronous belt transmission mechanism or a gear transmission mechanism.

[0019] Preferably, it also includes a control device, which is installed on the mobile frame, including a bracket installed on the mobile frame, a control module, a fingerprint module and an operation instruction input module arranged on the bracket, wherein the fingerprint module and the operation instruction input module are connected to the control module; the control module is connected to the infrared detection module, the camera module in the image acquisition device and the weighing module located on the detection platform.

[0020] A method for online detection of green tea appearance quality comprises the following steps:

[0021] S10, before starting to detect the green tea, the green tea needs to be calibrated by infrared and color, the thickness of the green tea is calculated by infrared calibration, and the error generated when extracting image features is reduced by color calibration;

[0022] S20, extracting the moisture content of green tea leaves, the ratio of buds to leaves, and the color, texture, and physical features of the image to construct a training set, a test set, and a validation set; wherein the following steps are included:

[0023] S21, obtaining image data of the green tea image, and ensuring that the green tea image fills the entire image and no background is visible;

[0024] S22, mixing one bud and two leaves and one bud and three leaves according to a certain proportion to form images with different bud and leaf proportions;

[0025] S23, using an electronic weighing instrument with an accuracy of 0.1g, taking an image of the green tea leaves every 5 minutes and recording the real-time weight information of the green tea leaves;

[0026] S24, after 10 hours of continuous water loss, the tea leaves are dried and weighed; assuming G 鲜 is the weight of green tea, G 烘 is the dry weight of the tea leaves after drying, W is the moisture content, t is the time, then the moisture content of the tea leaves at time t is:

[0027]

[0028] S25, reading the distance from the infrared detection module to the surface of the green tea at time t, and obtaining the thickness of the green tea at time t according to the distance calibrated at the beginning, and calculating the real-time density of the green tea in combination with the volume of the withering tank and the weight of the green tea at time t as the physical characteristic of the green tea;

[0029] S26, input a normal and complete green tea picture, extract the ROI region of interest and remove the border information;

[0030] S27, extract the feature values ​​of the six colors of the image, RGB, HSV, and Lab, in different color space domains;

[0031] S28. Establish the grayscale histogram of the image. According to the information of the histogram, the second-order moment, grayscale mean, consistency and entropy value can be obtained. Let z i represents the random variable of the grayscale of green tea, p represents the normalized histogram corresponding to green tea, l represents the maximum grayscale value, u represents the second-order moment, e represents the entropy value, and U represents the consistency. The texture feature can be expressed as:

[0032]

[0033]

[0034]

[0035]

[0036] S30, constructing a BP network model with multiple inputs and two outputs and multiple hidden layers, and reducing the dimension of the collected data information by using the dimension reduction characteristic of PCA;

[0037] S40, using GA genetic algorithm to train the BP network model to find the optimal region of the solution space and the number of dimensions of PCA dimensionality reduction, and finally determine its parameters;

[0038] S50, training the PCA-BP network model based on the weights and thresholds of the optimal individuals, and finally determining the parameters of the PCA-BP network model;

[0039] S60, inputting the color, texture and physical characteristic information of the green tea into the PCA-GA-BP network model, and outputting the moisture content and bud-leaf ratio.

[0040] Preferably, in step S30, PCA dimension reduction is performed on the data information to ensure that a small number of parameters contain relevant information of most parameters; a BP network model is constructed, y is set as the true value of water content, z is set as the predicted value of water content, E is set as the error, and the loss function is set as:

[0041]

[0042] The gradient descent method is used to propagate the error. Let w be the weight, b be the bias parameter, and η be the learning rate, which can be expressed as:

[0043]

[0044]

[0045] The activation function uses the relu function. Let a be the activated neuron and l be the number of network layers, which is expressed as:

[0046]

[0047] Among them, δ is expressed in the output layer as:

[0048]

[0049] The representation of δ in the hidden layer is:

[0050]

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] 1. The on-line detection device for the appearance quality of green tea leaves of the present invention can spread the green tea leaves to be detected into the withering trough on the detection platform, and has a higher degree of automation.

[0053] 2. The on-line detection device for the appearance quality of green tea leaves of the present invention can realize the weighing of the green tea leaves in the withering trough on the detection platform, and is used for real-time calculation of the weight of the green tea leaves in the withering trough on the detection platform.

[0054] 3. The on-line detection device for the appearance quality of green tea of ​​the present invention collects images of green tea through a camera module, and then calibrates the green tea through an infrared detection module to measure the thickness of the green tea in the withering trough laid flat on the detection platform.

[0055] 4. The online detection method for the appearance quality of green tea of ​​the present invention combines deep learning technology with machine vision technology, judges the appearance of tea leaves, extracts the moisture content of green tea, the ratio of buds and leaves, and the color, texture and physical characteristics of the image to construct a PCA-BP network model, and inputs the collected color, texture and physical characteristic information of green tea into the PCA-GA-BP network model to quickly obtain the moisture content and the ratio of buds and leaves, so as to judge the appearance quality of green tea. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1-Figure 2 Schematic diagrams of the three-dimensional structure from two different perspectives of the first specific embodiment of the online detection device for the appearance quality of green tea leaves of the present invention.

[0057] Figure 3-Figure 4 It is a schematic diagram of the three-dimensional structure of the tiling device, image acquisition device, control device and driving device from two different perspectives.

[0058] Figure 5-Figure 6 Schematic diagram of the three-dimensional structure of the tiling device from two different perspectives.

[0059] Figure 7 It is a simplified structural diagram of the tiling device.

[0060] Figure 8 The figure is a simplified structural diagram of the on-line detection device for the appearance quality of green tea leaves of the present invention.

[0061] Fig. 9 The figure is a schematic diagram of the process of the online detection method of the appearance quality of green tea leaves of the present invention.

[0062] Fig.10 This is the structural diagram of the PCA-GA-BP network model.

[0063] Fig.11 This is the flow chart of the PCA-GA-BP network model.

[0064] Fig.12 It is a schematic diagram of the three-dimensional structure of the driving device of the second specific embodiment of the on-line detection device for the appearance quality of green tea leaves of the present invention.

[0065] Fig.13 It is a schematic diagram of the three-dimensional structure of the storage mechanism of the third specific embodiment of the online detection device for the appearance quality of green tea leaves of the present invention.

[0066] Fig.14 for Fig.13 Schematic diagram of the structure of the storage mechanism.

[0067] Fig.15 It is a schematic diagram of the three-dimensional structure of the fourth specific embodiment of the online detection device for the appearance quality of green tea leaves of the present invention. DETAILED DESCRIPTION

[0068] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0069] Example 1

[0070] See also Figure 1-Figure 8 The on-line detection device for the appearance quality of green tea of ​​the present invention comprises a spreading device arranged on the detection platform 2 for spreading the green tea to be detected on the detection platform 2, an image acquisition device 4 for acquiring an image of the green tea after being spread on the detection platform 2, and a driving device 6 for driving the spreading device and the image acquisition device 4 to move along the length direction of the detection platform 2.

[0071] See also Figure 1-Figure 8 A withering trough 3 is provided on the detection platform 2, and a weighing module 404 is provided below the withering trough 3. The weighing module 404 is used to weigh the green tea leaves on the withering trough 3, such as an electronic scale or a pressure sensor.

[0072] See also Figure 1-Figure 8 The paving device comprises a storage mechanism 1 for holding the green tea to be tested and a green tea conveying mechanism arranged in the storage mechanism 1 for quantitatively conveying the green tea to the withering tank 3 on the testing platform 2; wherein,

[0073] The storage mechanism 1 is composed of a feeding part 101, a conveying part 103 and a flattening part 104, wherein the feeding part 101 is funnel-shaped, a conveying channel is provided in the conveying part 103, the upper end of the conveying channel is connected to the discharge port at the lower end of the feeding part 101, and the lower end is connected to one end of the flattening part 104; the other end of the flattening part 104 extends downward in a horizontal and inclined manner;

[0074] The green tea conveying mechanism includes a conveying roller 102 and a conveying motor for driving the conveying roller 102 to rotate, wherein the conveying roller 102 includes a rotating shaft and a plurality of partitions arranged on the rotating shaft, wherein the rotating shaft is arranged in the conveying part 103 and extends along the width direction of the conveying part 103; one end of the rotating shaft is rotatably connected to the conveying part 103, and the other end passes through the inside of the conveying part 103; the conveying part 103 is provided with a rotating bearing at the part in contact with the rotating shaft; a plurality of partitions are radially arranged on the rotating shaft and extend along the axial direction of the rotating shaft; the angles between two adjacent groups of partitions are equal; the conveying motor is installed on the outside of the conveying part 103 and connected to the rotating shaft; and in this embodiment, there are four groups of partitions, and the angles between two adjacent groups of partitions are 90 degrees; wherein the shape of the partition can be rectangular or arc-shaped, which is conducive to sending the green tea into the conveying channel.

[0075] Through the above arrangement, before the test, the tester first puts the green tea to be tested into the feeding part 101, and then, when the driving device 6 drives the spreading device to move along the length direction of the testing platform 2, the conveying motor drives the conveying roller 102 to rotate. While the conveying roller 102 rotates, the green tea between the two groups of partitions is sent to the conveying channel of the conveying part 103, and is evenly spread into the withering tank 3 of the testing platform 2 after passing through the spreading part 104 along the conveying channel. Since the angles between the two adjacent groups of partitions are equal, the green tea can be quantitatively conveyed to the spreading part 104 more evenly, so that the green tea is spread into the withering tank 3.

[0076] See also Figure 1-Figure 8 , the paving device also includes a lifting drive mechanism 7, which is arranged on the box body and is used to drive the storage mechanism 1 to perform lifting movement; the lifting drive mechanism 7 adopts a driving method combining a motor and a screw transmission mechanism, or adopts an electric push rod; through the above arrangement, when the paving device spreads the green tea on the above, the lifting drive mechanism 7 drives the storage mechanism 1 to rise, so that the paving device can spread the green tea on the green tea in the withering tank 3 again. In addition, since a sensor is provided at the bottom of the paving part 104, it can be ensured that the bottom of the paving part 104 will not hang on the green tea that has been laid flat, thereby ensuring that the paving work can proceed smoothly; and the sensor can adopt a non-contact sensor, such as an infrared sensor, to ensure that the bottom of the paving part 104 does not contact the green tea.

[0077] See also Figure 1-Figure 8The image acquisition device 4 includes a box, a camera module 401 arranged in the box, and an infrared detection module 402, wherein the camera module 401 includes a camera bracket and a camera; the infrared detection module 402 can use an infrared detection probe, and a light source 403 can also be arranged in the box to increase the brightness.

[0078] See also Figure 1-Figure 8 The driving device 6 includes a moving frame 601 and a moving driving mechanism for driving the moving frame 601 to move, wherein two groups of walking wheels 602 are arranged on the moving frame 601, and the two groups of walking wheels 602 are respectively arranged on the front and rear sides of the moving frame 601, and each group of walking wheels 602 has two, and the two walking wheels 602 are respectively arranged on the left and right sides of the moving frame 601, and each walking wheel 602 can also be provided with a group of moving motors 603 separately, which are used to drive the walking wheel 602 to move separately; the walking wheel 602 is driven to move by the moving motor 603, thereby driving the tiling device and the image acquisition device 4 to reciprocate back and forth along the length direction of the detection platform 2; in addition, the walking wheel 602 is provided with a groove 604 that cooperates with the edge of the detection platform 2, and the groove 604 can constitute a guide groove for guiding the linear motion of the moving frame 601, so as to improve the motion accuracy.

[0079] See also Figure 1-Figure 8 The on-line detection device for the appearance quality of green tea of ​​the present invention also includes a control device 5, which is installed on the mobile frame 601, and includes a bracket installed on the mobile frame 601, a control module, a fingerprint module and an operation instruction input module arranged on the bracket, wherein the fingerprint module and the operation instruction input module are connected to the control module; the control module is connected to the infrared detection module, the camera module in the image acquisition device 4 and the weighing module 404 located on the detection platform 2.

[0080] See also Figure 9-11 The process of the on-line detection method for the appearance quality of green tea of ​​the present invention is as follows:

[0081] S10, before the on-line detection device for the appearance quality of green tea of ​​the present invention starts to detect green tea, infrared calibration and color calibration are required;

[0082] S20, extracting the moisture content of green tea leaves, the ratio of buds to leaves, and the color, texture, and physical features of the image to construct a training set, a test set, and a validation set;

[0083] S30, constructing a BP model with multiple hidden layers and multiple inputs and two outputs, and using the dimension reduction characteristics of PCA to reduce the dimension of a large amount of information, so that more image information can be expressed with fewer parameters, thereby increasing the robustness of the model;

[0084] S40, using GA genetic algorithm to train the BP network to find the optimal area of ​​the solution space and the number of dimensions of PCA dimensionality reduction, and finally determine its parameters;

[0085] S50, training the PCA-BP network based on the weights and thresholds of the optimal individuals, and finally determining the parameters of the model;

[0086] S60, inputting the color, texture and physical characteristic information of the green tea into the PCA-GA-BP model, and outputting the moisture content and bud-leaf ratio;

[0087] Among them, infrared calibration and color calibration are required before starting to detect the tea leaves. The purpose of infrared calibration is to accurately calculate the thickness of the tea leaves, and the purpose of color calibration is to reduce the error generated when extracting image features.

[0088] In the process of extracting the moisture content of green tea leaves and the color and texture features of the image to construct the training set, test set, and validation set, the image data of the green tea leaves is first obtained, and it must be ensured that the green tea leaves fill the entire image and no background is visible. An electronic weighing instrument with an accuracy of 0.1g is used to take a green tea leaf image every 5 minutes while recording the real-time weight information of the green tea leaves. After 10 hours of continuous water loss, the green tea leaves are dried and weighed. Assuming G 鲜 is the weight of green tea, G 烘 is the dry weight of the tea leaves after drying, W is the moisture content, t is the time, then:

[0089]

[0090] Read the distance from the infrared probe to the surface of the green tea at time t. Based on the distance calibrated at the beginning, the thickness of the green tea at time t can be obtained. Since the volume of the withering trough 3 is known and the weight at time t is known, the real-time density of the green tea can be calculated as the physical characteristic of the green tea.

[0091] Input a normal and complete tea green picture, extract the ROI region of interest and remove the border information. Extract the eigenvalues ​​of the six colors of the image, RGB, HSV, and Lab, in different color space domains. Build the grayscale histogram of the image. Based on the information in the histogram, we can derive the second-order moment (standard), grayscale mean (mean), consistency (consistency) and entropy (entropy). Let z i represents the random variable of the grayscale of green tea, p represents the normalized histogram corresponding to green tea, l represents the maximum grayscale value, u represents the second-order moment, e represents the entropy value, and U represents the consistency. The texture feature can be expressed as:

[0092]

[0093]

[0094]

[0095]

[0096] Construct a BP model with multiple hidden layers and multiple inputs and one output. Use the dimension reduction characteristics of PCA to reduce the dimension of a large amount of information, thereby increasing the robustness of the model, including: performing PCA dimension reduction on the data to ensure that a small number of parameters contain relevant information about most parameters. Construct a BP network, set the true value of y moisture content, the predicted value of z moisture content, E as the error, and set the loss function as:

[0097]

[0098] The gradient descent method is used to propagate the error. Let w be the weight, b be the bias parameter, and η be the learning rate, which can be expressed as:

[0099]

[0100]

[0101] The activation function uses the relu function to add nonlinear properties to the neuron to make it more robust. Let a be the activated neuron and l be the number of network layers, which can be expressed as:

[0102]

[0103] Among them, δ is expressed in the output layer as:

[0104]

[0105] The representation of δ in the hidden layer is:

[0106]

[0107] The GA genetic algorithm trains the BP network to find the optimal area of ​​the solution space and the number of dimensions of PCA dimensionality reduction, and finally determines its parameters, including: BP threshold, PCA dimension number, and encoding of the topological structure information of the BP network as the optimization object of the GA algorithm. The number of samples, number of chromosomes, and probability of crossover mutation of the GA algorithm are constructed to calculate the fitness function A according to the BP loss function E:

[0108]

[0109] The worst individual is selected and replaced with the best individual of the previous generation to ensure the convergence of the GA algorithm.

[0110] After finally calculating the optimal threshold and weight, the number of neurons and the number of PCA, the texture information and color information of the image are input to predict the moisture content of the green tea.

[0111] Example 2

[0112] See also Fig.12 The difference between this embodiment and embodiment 1 is that: the left and right walking wheels 602 located at the front side and / or the rear side are connected by a connecting shaft 606, and the front and rear walking wheels 602 located at the left and right sides are connected by a synchronous belt transmission mechanism 603; the mobile drive mechanism includes a mobile motor, and the mobile motor is installed on the mobile frame 601, and the output shaft of the mobile motor is directly connected to one of the connecting shafts, or is connected to one of the connecting shafts through a synchronous belt transmission mechanism or a gear transmission mechanism.

[0113] The remaining structures are implemented with reference to Example 1.

[0114] Example 3

[0115] See also Figure 13-14 , the difference between this embodiment and embodiment 1 is that: both sides of the conveying channel of the vertical portion 103 are provided with air holes 106, and the air holes 106 are multiple and arranged in an array; the multiple air holes 106 are connected to the air pump through the air pipe. By setting the above-mentioned air holes 106 to blow air into the conveying channel, it is possible to prevent the green tea from sticking to the inner wall of the conveying channel, wherein 105 is the outlet of the storage mechanism in this embodiment, and the direction of the outlet 105 can be located on the same side as the image acquisition device, that is, toward the image acquisition device, and the mobile frame can also be provided with an avoidance groove for the green tea in the outlet to pass through, or the height of the outlet of the storage mechanism can be set below the mobile frame, so that in the process of paving the green tea, the image acquisition device located at the rear can capture the image of the paved green tea, without waiting for the green tea to be paved, and the driving device drives the image acquisition device to move in the opposite direction to collect the green tea data, which is conducive to improving the image acquisition accuracy.

[0116] The remaining structures are implemented with reference to Example 1.

[0117] Example 4

[0118] See also Fig.15 The difference between this embodiment and embodiment 1 is that this embodiment combines the differences between embodiments 2 and 3.

[0119] The above is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited to the above content. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. An online detection device for the appearance quality of green tea, characterized in that: The invention comprises a spreading device arranged on the detection platform for spreading the green tea to be detected on the detection platform, an image acquisition device for acquiring images of the green tea spread on the detection platform, and a driving device for driving the spreading device and the image acquisition device to move along the length direction of the detection platform, wherein: The detection platform is provided with a withering trough, and a weighing module is provided below the withering trough, and the weighing module is used to weigh the green tea leaves on the withering trough; The paving device comprises a storage mechanism for holding the green tea to be tested and a green tea conveying mechanism arranged in the storage mechanism for quantitatively conveying the green tea to the withering tank on the testing platform; The image acquisition device includes a box, a camera module and an infrared detection module arranged in the box; The storage mechanism is composed of a feeding part, a conveying part and a flattening part, wherein the feeding part is funnel-shaped, a conveying channel is arranged in the conveying part, the upper end of the conveying channel is connected with the discharge port at the lower end of the feeding part, and the lower end is connected with one end of the flattening part; the other end of the flattening part extends downwardly in a horizontal manner; Both sides of the conveying channel of the conveying part are provided with air holes, and the air holes are multiple and arranged in an array; the multiple air holes are connected to the air pump through the air pipe; A sensor is provided at the bottom of the flattening portion; The green tea conveying mechanism comprises a conveying roller and a conveying motor for driving the conveying roller to rotate, wherein the conveying roller comprises a rotating shaft and a plurality of partitions arranged on the rotating shaft, wherein the rotating shaft is arranged in the conveying part and extends along the width direction of the conveying part; one end of the rotating shaft is rotatably connected to the conveying part, and the other end passes through the inside of the conveying part; the conveying part is provided with a rotating bearing at a position in contact with the rotating shaft; a plurality of partitions are radially arranged on the rotating shaft and extend along the axial direction of the rotating shaft; the angles between two adjacent groups of partitions are equal; the conveying motor is installed outside the conveying part and connected to the rotating shaft; The paving device also includes a lifting drive mechanism, which is arranged on the box body and is used to drive the storage mechanism to perform lifting movements; the lifting drive mechanism adopts a driving method combining a motor and a screw transmission mechanism, or adopts an electric push rod.

2. The on-line detection device for the appearance quality of green tea leaves according to claim 1 is characterized in that: The driving device includes a mobile frame and a mobile driving mechanism for driving the mobile frame to move, wherein the mobile frame is provided with two groups of running wheels, the two groups of running wheels are respectively arranged on the front and rear sides of the mobile frame, each group of running wheels has two running wheels, and the two running wheels are respectively arranged on the left and right sides of the mobile frame; wherein the running wheels on the front and rear sides are connected by a synchronous belt transmission mechanism, and the running wheels on the left and right sides are connected by a connecting shaft; the mobile driving mechanism includes a mobile motor, the mobile motor is installed on the mobile frame, and the output shaft of the mobile motor is connected to one of the connecting shafts through a synchronous belt transmission mechanism or a gear transmission mechanism.

3. The on-line detection device for the appearance quality of green tea according to claim 2 is characterized in that: It also includes a control device, which is installed on the mobile frame, including a bracket installed on the mobile frame, a control module, a fingerprint module and an operation instruction input module arranged on the bracket, wherein the fingerprint module and the operation instruction input module are connected to the control module; the control module is connected to the infrared detection module, the camera module in the image acquisition device and the weighing module located on the detection platform.

4. A method for online detection of green tea appearance quality used in the device for online detection of green tea appearance quality according to any one of claims 1 to 3, characterized in that: The following steps are involved: S10, before starting to detect the green tea, the green tea needs to be calibrated by infrared and color, the thickness of the green tea is calculated by infrared calibration, and the error generated when extracting image features is reduced by color calibration; S20, extracting the moisture content of green tea leaves, the ratio of buds to leaves, and the color, texture, and physical features of the image to construct a training set, a test set, and a validation set; The steps include: S21, obtaining image data of the green tea image, and ensuring that the green tea image fills the entire image and no background is visible; S22, mixing one bud and two leaves and one bud and three leaves according to a certain proportion to form images with different bud and leaf proportions; S23, using an electronic weighing instrument with an accuracy of 0.1g, taking an image of the green tea leaves every 5 minutes and recording the real-time weight information of the green tea leaves; S24. After the water loss lasts for 10 hours, the green tea is dried and weighed; assuming that G fresh is the weight of the green tea, G roasted is the dry weight of the green tea after drying, W represents the moisture content, t represents the time, then the moisture content of the green tea at time t is: ； S25, reading the distance from the infrared detection module to the surface of the green tea at time t, and obtaining the thickness of the green tea at time t according to the distance calibrated at the beginning, and calculating the real-time density of the green tea in combination with the volume of the withering tank and the weight of the green tea at time t as the physical characteristic of the green tea; S26, input a normal and complete green tea picture, extract the ROI region of interest and remove the border information; S27, extract the feature values ​​of the six colors of the image, RGB, HSV, and Lab, in different color space domains; S28. Establish the grayscale histogram of the image, and obtain the second-order moment, grayscale mean, consistency and entropy value based on the information of the grayscale histogram. Let zi represent the random variable of the grayscale of green tea, p represent the normalized histogram corresponding to green tea, l represent the maximum grayscale value, u represent the second-order moment, e represent the entropy value, and U represent the consistency. The texture feature is expressed as: ； ； ； ； S30, constructing a BP network model with multiple inputs and two outputs and multiple hidden layers, and reducing the dimension of the collected data information by using the dimension reduction characteristic of PCA; S40, using GA genetic algorithm to train the BP network model to find the optimal region of the solution space and the number of dimensions of PCA dimensionality reduction, and finally determine its parameters; S50, training the PCA-BP network model based on the weights and thresholds of the optimal individuals, and finally determining the parameters of the PCA-BP network model; S60, inputting the color, texture and physical characteristic information of the green tea into the PCA-GA-BP network model, and outputting the moisture content and bud-leaf ratio.

5. The method for online detection of green tea appearance quality according to claim 4, characterized in that: In step S30, PCA dimension reduction is performed on the data information to ensure that a small number of parameters contain relevant information of most parameters; a BP network model is constructed, y is set as the true value of water content, z is the predicted value of water content, E is the error, and the loss function is set as: ； The gradient descent method is used to propagate the error. Let w be the weight, b be the bias parameter, and η be the learning rate, which can be expressed as: ； ； The activation function uses the relu function. Let a be the activated neuron and l be the number of network layers, which is expressed as: ； Among them, δ is expressed in the output layer as: ； The representation of δ in the hidden layer is: 。

Images