Tea production line flow intelligent detection method, equipment, medium and product

Abstract

The invention discloses an intelligent flow detection method and device for a tea production line, a medium and a product, and relates to the technical field of tea processing. The prediction model is a multi-task learning model constructed and optimized based on a CNN model, a full connection layer and an optimizer; obtaining two-dimensional image data of the tea leaves in real time; inputting the two-dimensional image data of the tea leaves into a prediction model to obtain a prediction process type, a prediction conveyor belt speed and a prediction weight of the tea leaves; obtaining a tea passing distance based on the tea two-dimensional image data; and determining the tea flow based on the predicted tea weight, the tea passing distance and the conveyor belt speed. According to the invention, high-precision prediction and real-time monitoring of the tea flow can be realized.

Description

Technical Field

[0001] The present application relates to the technical field of tea processing, and in particular to a method, equipment, medium and product for intelligently detecting flow of a tea production line. Background Art

[0002] In the primary processing and refining process of tea, the balanced and stable flow of tea is the key factor to ensure the quality of tea. For example, in the drum withering (primary processing) of green tea, the stable flow of tea corresponds to the fixed withering temperature and drum speed. If the flow of tea changes, it is necessary to adjust the set temperature or drum speed of the drum withering machine in real time to ensure the consistent quality of tea withering. For the automatic withering machine, the system controller is required to be able to adjust the withering process parameters (mainly the set temperature and set speed of the drum withering machine) in real time according to the changes in the tea flow. For the manually controlled withering machine, if the real-time flow of tea can be known, the withering process parameters can be accurately adjusted. Therefore, for the withering process of tea, whether it is automatic control or manual control, the first prerequisite for ensuring quality is the need to detect the flow of tea in real time. This principle also applies to other primary processing links of tea such as drying and rolling, and refining links such as screening and air selection. In addition, the real-time detection of tea flow can further realize the quantitative feeding of tea. In the tea production process, traditional flow detection methods mainly rely on mechanical weighing or flow meter equipment, which have problems such as poor real-time performance, low accuracy and being greatly affected by vibration. Summary of the invention

[0003] The purpose of this application is to provide a tea production line flow intelligent detection method, equipment, medium and product, which can achieve high-precision prediction and real-time monitoring of tea flow, and then accurately adjust the processing parameters and quantitative feeding of tea, so as to improve the quality of tea processing.

[0004] To achieve the above objectives, this application provides the following solutions: In a first aspect, the present application provides a method for intelligent flow detection of a tea production line, comprising: Obtain a prediction model; the prediction model is a multi-task learning model constructed and optimized based on a CNN model, a fully connected layer, and an optimizer; Acquire tea leaves 2D image data in real time; Inputting the two-dimensional image data of tea leaves into the prediction model to obtain the predicted process type, the predicted conveyor belt speed and the predicted weight of tea leaves; Obtaining the distance traveled by the tea leaves based on the two-dimensional image data of the tea leaves; The tea flow rate is determined based on the predicted weight of the tea leaves, the distance the tea leaves have traveled, and the conveyor belt speed.

[0005] Optionally, the construction process of the prediction model includes: Obtain sample tea leaf image data, process labels, conveyor belt speed labels, and sample tea leaf actual weight data; the sample tea leaf image data includes sample tea leaf two-dimensional image data and sample tea leaf three-dimensional image data; Based on the sample tea leaf three-dimensional image data, obtain the actual volume data of the sample tea leaf; Based on the process labels, the conveyor belt speed labels, the actual volume data of the sample tea leaf, and the actual weight data of the sample tea leaf, construct a sample data set; Based on the CNN model, the fully connected layer, and the optimizer, construct an initial prediction model; Input the sample tea leaf two-dimensional image data into the initial prediction model to obtain the predicted process type of the sample tea leaf, the predicted conveyor belt speed, and the predicted weight data of the sample tea leaf; Train the initial prediction model based on the sample data set until the output result of the trained initial prediction model meets the set requirements, and use the trained initial prediction model as the prediction model.

[0006] Optionally, obtaining the actual volume data of the sample tea leaf based on the sample tea leaf three-dimensional image data includes: Calibrate the sample tea leaf three-dimensional image to obtain calibrated image data; Based on the calibrated image data, obtain point cloud data; Based on the point cloud data, obtain the actual volume data of the sample tea leaf.

[0007] Optionally, the actual weight data of the sample tea leaf corresponds one-to-one with the actual volume data of the sample tea leaf; Based on the actual volume data of the sample tea leaf and the corresponding actual weight data of the sample tea leaf, construct a sample data set.

[0008] Optionally, inputting the sample tea leaf two-dimensional image data into the initial prediction model to obtain the predicted process type of the sample tea leaf, the predicted conveyor belt speed, and the predicted weight data of the sample tea leaf includes: Normalize and tensorize the sample tea leaf two-dimensional image data to obtain processed image data; Based on the processed image data, use the initial prediction model to obtain the predicted process type of the sample tea leaf, the predicted conveyor belt speed, and the predicted weight data of the sample tea leaf.

[0009] Optionally, based on the processed image data, using the initial prediction model to obtain the predicted process type of the sample tea leaf, the predicted conveyor belt speed, and the predicted weight data of the sample tea leaf includes: Based on the processed image data, a shared feature is obtained using a CNN model; Based on the shared feature, a predicted process type of the sample tea leaves and a predicted conveyor belt speed are obtained using a fully connected layer; Based on the shared feature, predicted volume data of the sample tea leaves is obtained using the fully connected layer; Based on the shared feature and the predicted volume data of the sample tea leaves, predicted weight data of the sample tea leaves is obtained using the fully connected layer.

[0010] Optionally, an adaptive joint loss function is used to train the initial prediction model based on the sample data set; the adaptive joint loss function includes a mean square error loss function.

[0011] In a second aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the steps of the tea production line flow intelligent detection method described in any one of the above.

[0012] In a third aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the tea production line flow intelligent detection method described in any one of the above are implemented.

[0013] In a fourth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the tea production line flow intelligent detection method described in any one of the above are implemented.

[0014] According to the specific embodiments provided by the present application, the present application has the following technical effects: The present application provides a tea production line flow intelligent detection method, device, medium, and product. By constructing a multi-task learning model based on a CNN model, a fully connected layer, and an optimizer as a prediction model, the process type to which the tea belongs, the speed of the conveyor belt, and the weight of the tea are predicted according to the two-dimensional image data of the tea obtained in real time. The distance passed by the tea is determined according to the two-dimensional image of the tea, and the tea flow is determined according to the predicted tea weight, the distance passed by the tea, and the conveyor belt speed, realizing high-precision prediction and real-time monitoring of the tea flow, and further realizing precise adjustment of the processing process parameters and quantitative feeding of the tea according to the predicted process type and tea flow, improving the tea processing quality. Description of the Drawings

[0015] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0016] Figure 1 It is a flowchart of an intelligent detection method for the flow rate of a tea production line in an embodiment of the present application; Figure 2 It is a schematic diagram of the flow rate prediction process of the intelligent detection method for the flow rate of a tea production line provided in an embodiment of the present application; Figure 3 It is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. Detailed implementation manners

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0018] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0019] In an exemplary embodiment, as Figure 1 shown, an intelligent detection method for the flow rate of a tea production line is provided. This method is executed by a computer device, which can be specifically executed by a computer device such as a terminal or a server alone, or jointly executed by a terminal and a server. In the embodiments of the present application, taking the prediction of the tea flow rate by this method as an example, it includes: Step 100, obtain a prediction model. The prediction model is a multi-task learning model constructed and optimized based on a CNN model, a fully connected layer, and an optimizer.

[0020] Step 200, obtain tea two-dimensional image data in real time.

[0021] Step 300, input the tea two-dimensional image data into the prediction model to obtain a predicted process type, a predicted conveyor belt speed, and a predicted tea weight.

[0022] Step 400, obtain the distance passed by the tea based on the tea two-dimensional image data. Determine the tea flow rate based on the predicted tea weight, the distance passed by the tea, and the conveyor belt speed.

[0023] It should be noted that in the actual application process, the industrial camera installed directly above the tea conveyor belt takes vertical downward shots to collect two-dimensional tea image data.

[0024] As an alternative implementation, to improve the accuracy of tea flow prediction, step 100 includes: 101. Obtain sample tea image data, process labels, conveyor belt speed labels, and actual sample tea weight data. The sample tea image data includes two-dimensional sample tea image data and three-dimensional sample tea image data.

[0025] In the actual acquisition of sample tea image data and sample tea actual weight data, an industrial camera and a depth camera are installed directly above the tea conveyor belt and shoot vertically downward. The industrial camera collects two-dimensional images, and the depth camera collects three-dimensional images. Among them, the industrial camera can use the Hikvision MV-CS060-10GC camera, and the depth camera can use the Microsoft Azure Kinect DK depth camera (including an RGB camera and a depth camera, simply referred to as the Kinect camera in all embodiments provided in this application) to ensure the clarity and integrity of the acquired tea images. The acquisition method of the sample tea image data is as follows: Samples (obtaining process labels) after the completion of the fixation, rolling, and primary drying processes are collected respectively on the green tea production line of a certain tea production company. Use a Bergso small conveyor and a self-made conveyor belt to simulate the actual production situation, and cooperate with a stepping motor to control the conveyor belt speed. Set the conveyor belt speed to three speeds of 0.1 m / s, 0.2 m / s, and 0.3 m / s (obtaining conveyor belt speed labels), and set the conveyor speed to 0.2 m / s. Evenly scatter the tea on the conveyor, start the conveyor belt, and the tea falls from the upper conveyor to the conveyor belt below. The weight of each batch of tea samples is between 0 and 15 g, and there is a certain distance between batches to ensure that there is only one batch of samples in the camera shooting area each time the image data is acquired. When all the tea enters the camera shooting range, the conveyor belt below stops, and the Kinect camera and the industrial camera shoot images to obtain sample tea images. An electronic scale records the weight of the tea falling on the conveyor belt to obtain the actual weight of the sample tea, which corresponds one by one to the images taken by the camera. Sequentially change the conveyor belt speed and correspond to the tea samples of different processes, and repeat the above process for data acquisition to obtain sample tea image data and sample tea actual weight data. In the embodiments provided in this application, a total of 1000 samples are collected in the fixation process, including 200 samples with a conveyor belt speed of 0.1 m / s, 600 samples with a conveyor belt speed of 0.2 m / s, and 200 samples with a conveyor belt speed of 0.3 m / s; a total of 1000 samples are collected in the rolling process, including 200 samples with a conveyor belt speed of 0.1 m / s, 600 samples with a conveyor belt speed of 0.2 m / s, and 200 samples with a conveyor belt speed of 0.3 m / s; a total of 1000 samples are collected in the primary drying process, including 200 samples with a conveyor belt speed of 0.1 m / s, 200 samples with a conveyor belt speed of 0.2 m / s, and 600 samples with a conveyor belt speed of 0.3 m / s. Due to problems such as equipment failures and manual operations, data loss and abnormal data will inevitably occur. In this embodiment, the processing method of removing the data samples containing missing values is adopted, and there are 2689 samples after processing.

[0026] 102. Obtain the actual volume data of the sample tea leaves based on the three-dimensional image data of the sample tea leaves. Construct a sample data set based on the actual volume data and the actual weight data of the sample tea leaves. This includes: calibrating the three-dimensional image of the sample tea leaves to obtain calibrated image data. Obtain point cloud data based on the calibrated image data. Obtain the actual volume data of the sample tea leaves based on the point cloud data.

[0027] For example, based on the actual acquisition process of the above sample tea leaf image data and the actual weight data of the sample tea leaves, calibrate the Kinect camera. Each time the Kinect camera takes a picture, it will simultaneously capture an RGB image and a depth image. Use the RGB camera and the depth camera to take 20 calibration pictures of the checkerboard at different angles and positions. The checkerboard specifications are a board size of 100mm × 100mm, a grid side length of 10mm, and a pattern size of 90mm × 90mm (i.e., 9 × 9 grids). Use OpenCV to read the captured calibration images, and return the internal parameter matrix K, distortion coefficient D, external parameter matrix RT (rotation matrix R and translation vector T), and reprojection error of the RGB camera and the depth camera through relevant functions.

[0028] For each pixel point in the depth image , according to the depth value and the internal parameters of the depth camera calculate the three-dimensional coordinates of the pixel point in the depth camera coordinate system: , , . In the formula, is the coordinate of the pixel point in the depth camera coordinate system, is the depth value of the pixel point in the depth image, is the focal length of the depth camera, is the principal point coordinate of the depth image.

[0029] Use the external parameter matrix RT to transform the pixel point from the depth camera coordinate system to the RGB coordinate system: . In the formula, represents the three-dimensional coordinate of the pixel point in the depth image in the RGB coordinate system.

[0030] Use the internal parameters of the RGB camera to project the three-dimensional coordinate of the pixel point in the RGB coordinate system onto the RGB image plane: , , . In the formula, represents the pixel point coordinate projected onto the RGB image, is the focal length of the RGB camera, Represents the main point coordinates of the RGB image. Thus, the projected RGB image is obtained. The projected RGB image is converted from the RGB color space to the HSV color space. The tea leaf area in the image is extracted through the set color threshold to generate a binary mask. Through the closing operation in morphological operations, a 5×5 matrix is used as the structuring element to process the mask, smoothing the object boundaries in the image. The RGB image is registered with the depth image. Combining the above binary mask, the depth data of the tea leaf area in the depth image is extracted. The three-dimensional coordinates of each pixel point are calculated according to the internal parameters and depth values of the depth camera to obtain the point cloud data. The calculation method of the three-dimensional coordinates refers to the process of calculating the three-dimensional coordinates of the pixel points in the depth image in the depth camera coordinate system above. After obtaining the point cloud data, the ConvexHull algorithm (convex hull algorithm) is used to calculate the outer bounding volume of the point cloud, thereby calculating the tea leaf volume and obtaining the actual volume data of the sample tea leaves. The actual weight data of the sample tea leaves corresponds one-to-one with the actual volume data of the sample tea leaves. A sample dataset is constructed based on the process label, conveyor belt speed label, actual volume data of the sample tea leaves, and actual weight data of the sample tea leaves. To evaluate and optimize the model performance, all the sample data in the sample dataset are randomly divided into a training dataset and a test dataset in a ratio of 4:1.

[0031] 103. An initial prediction model is constructed based on the CNN model, fully connected layer, and optimizer.

[0032] For example, ResNet18 is used as the backbone network (i.e., the CNN model) to extract features from the image input. Each residual block in the CNN model includes a normalization layer and a Rectified Linear Unit (ReLU) as the activation function. These residual blocks extract latent features from the input image data, convert them into 512-dimensional features, and feed them into the fully connected layer. Among them, the fully connected layer can be a 2-layer Multilayer Perceptron (MLP), with ReLU added in the middle as the activation function, outputting 32-dimensional shared features. To avoid overfitting, a dropout layer is added in the middle of the fully connected layer. The optimizer optimizes the fully connected layer during the training of the Multi-Task Learning (MTL) model.

[0033] 104. Input the two-dimensional image data of the sample tea leaves into the initial prediction model to obtain the predicted process type of the sample tea leaves, the predicted conveyor belt speed, and the predicted weight data of the sample tea leaves. This includes: normalizing and tensorizing the two-dimensional image data of the sample tea leaves to obtain the processed image data. Based on the processed image data, use the initial prediction model to obtain the predicted process type of the sample tea leaves, the predicted conveyor belt speed, and the predicted weight data of the sample tea leaves. Among them, based on the processed image data, use the CNN model to obtain the shared features. Based on the shared features, use the fully connected layer to obtain the predicted process type of the sample tea leaves, the predicted conveyor belt speed, and the predicted volume data of the sample tea leaves. Based on the shared features and the predicted volume data of the sample tea leaves, use the fully connected layer to obtain the predicted weight data of the sample tea leaves.

[0034] For example, preprocess the two-dimensional image data of the sample tea leaves. Downsample the image. The original image resolution is 3072×2048, and it is scaled to 1024×768 pixels. Then perform normalization and tensorization processing. The normalization mean is (0.485, 0.456, 0.406), and the standard deviation is (0.229, 0.224, 0.225) to obtain the processed image data. As Figure 2 shown, input the processed image data into the initial prediction model. The initial prediction model extracts potential features from the input data through the convolutional layer and the residual block. After passing through the convolutional layer, it is converted into 512-dimensional features through the global average pooling layer and fed into the fully connected layer; the fully connected layer outputs 32-dimensional shared features; in order to avoid overfitting, a dropout layer is added in the middle of the fully connected layer. The neural network includes four different output layers: process classification, conveyor belt speed classification, volume regression, and weight regression. The extracted shared features pass through the fully connected layer and output the probability distributions of three types of processes and three types of speeds. The output of the classification task is a three-dimensional vector, corresponding to the three process types of fixation, rolling, and primary drying and the three conveyor belt speeds of 0.1m / s, 0.2m / s, and 0.3m / s respectively. The extracted shared features pass through the fully connected layer to output the volume value. The extracted shared features are concatenated with the volume value and input into the fully connected layer to output the weight value. The output layer of the regression task is a one-dimensional vector, corresponding to volume and weight respectively. Each output is returned through the ReLU function and the sigmoid function, and finally the predicted weight data of the sample tea leaves is obtained.

[0035] 105. Train the initial prediction model based on the sample data set until the output result of the trained initial prediction model meets the set requirements, and then use the trained initial prediction model as the prediction model. Among them, use the adaptive joint loss function to train the initial prediction model based on the sample data set. The adaptive joint loss function includes the cross-entropy loss function and the mean squared error loss function.

[0036] For example, to optimize the multi-task learning model, this application designs an adaptive joint loss function (Adaloss, AL) to enable collaborative optimization of the volume prediction and weight prediction tasks. The mean squared error loss function is used as the loss function for the regression task, expressed as: , . In the formula, represents the total loss function, represents the loss of the th regression task, is the learnable task weight parameter; represents the normalization factor, which is used to ensure that the sum of the weights is 1; is used to dynamically adjust the task weights. The specific steps of the adaptive joint loss function are as follows: First, initialize the learnable parameters , with one weight parameter corresponding to each task; Next, calculate the individual loss , ; Normalize the weights using the Softmax function; Finally, calculate the joint loss. Among them, the weight parameters are trained and optimized using the backpropagation algorithm. The multi-task learning model uses Adaptive Moment Estimation (Adam) as the optimizer to train the initial prediction model, and the grid search method is used to determine that the learning rate, mini-batch, and epoch size of the optimizer are , 32, and 310 respectively. Set the sequential loss switching strategy as follows: The first 10 epochs are used for the classification task, the next 100 epochs are used for the volume regression task, and the last 200 epochs are used for the weight regression task. When the output result of the initial prediction model after training reaches the set requirements, the trained initial prediction model is used as the prediction model. The Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and correlation coefficient R are used as evaluation indicators to determine the performance of the model. The smaller the RMSE and MAE, the more accurate the prediction result. The correlation coefficient R represents the fitting degree between the true value and the predicted value, and the closer R is to 1, the better the fitting effect.

[0037] Input the sample data in the sample dataset into the initial prediction model to obtain the predicted process type, the predicted conveyor belt speed, and the predicted weight data of the sample tea leaves, and train the initial prediction model. When the output result of the trained initial prediction model meets the set requirements, use the trained initial prediction model as the prediction model. Among them, input the sample data in the sample dataset into the initial prediction model to obtain the predicted process type and the predicted conveyor belt speed, and train the classification task in the initial prediction model. Input the sample data in the sample dataset into the initial prediction model, obtain 32-dimensional shared features through the CNN model, obtain the predicted volume data of the sample tea leaves through the fully connected layer based on the shared features, and splice the shared features with the predicted volume data of the sample tea leaves. Based on the spliced result, use the fully connected layer to obtain the predicted weight data of the sample tea leaves, and dynamically adjust the weights of the two regression tasks through the adaptive loss function to train the volume regression task and the weight regression task in the initial prediction model.

[0038] Step 400 includes: obtaining the distance passed by the tea leaves based on the two-dimensional image data of the tea leaves. Based on the predicted weight of the tea leaves, the distance passed by the tea leaves, and the conveyor belt speed, use the formula to determine the tea leaf flow rate. In the formula, represents the tea leaf flow rate, represents the predicted weight of the tea leaves, represents the conveyor belt speed, represents the distance passed by the tea leaves.

[0039] In an exemplary embodiment, the prediction model provided by the present application, the conventional ResNet18 model, and the multi-task learning model without an adaptive loss function are respectively used for prediction, and the evaluation indexes of the prediction results are shown in Table 1.

[0040] Table 1 Comparison table of prediction result evaluation indexes

[0041] It can be seen from Table 1 that compared with the conventional ResNet18 model and the multi-task learning model without an adaptive loss function, the prediction model in the present application has higher accuracy, and the value in the mean absolute percentage error is the smallest, which fully shows that the intelligent detection method for the tea production line flow proposed in the present application has better improvement in terms of prediction accuracy and balance.

[0042] Combined with the above embodiments, the present application can achieve the following beneficial effects: Based on the industrial camera, the present application uses machine vision technology to realize real-time, continuous, and accurate measurement of the tea leaf flow rate on the conveyor belt. By establishing the relationship between the image and the volume, and the volume and the weight, the real-time tea leaf weight information is obtained, and the tea leaf flow rate is obtained in combination with the conveyor belt speed.

[0043] This application can establish the mapping relationship from RGB images to volume and then to weight, effectively combining multi-modal features to ensure the efficiency and applicability of the model.

[0044] This application can achieve the automated monitoring of tea flow, reduce the dependence on manual inspection, significantly improve the automation level and operation efficiency of the production line. During the production process, the conveyor belt speed and feeding volume can be dynamically adjusted according to the flow prediction results to optimize the production rhythm and avoid production delays or resource waste caused by excessive or insufficient feeding.

[0045] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as Figure 3 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store relevant data involved in the intelligent detection method of tea production line flow. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements an intelligent detection method for tea production line flow.

[0046] Those skilled in the art can understand that Figure 3 the structure shown in

[0047] is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0048] In an exemplary embodiment, a computer program product is provided, including a computer program which, when executed by a processor, implements the steps in the above method embodiments.

[0049] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0050] Those of ordinary skill in the art can understand that all or part of the processes of implementing the above method embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above method embodiments. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0051] The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0052] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0053] In this text, specific examples are used to elaborate on the principles and implementation manners of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A tea production line flow intelligent detection method, characterized in that: The tea production line flow intelligent detection method comprises: Obtain a prediction model; the prediction model is a multi-task learning model constructed and optimized based on a CNN model, a fully connected layer, and an optimizer; Acquire tea leaves 2D image data in real time; Inputting the two-dimensional image data of tea leaves into the prediction model to obtain the predicted process type, the predicted conveyor belt speed and the predicted weight of tea leaves; Obtaining the distance traveled by the tea leaves based on the two-dimensional image data of the tea leaves; The tea flow rate is determined based on the predicted weight of the tea leaves, the distance the tea leaves have traveled, and the conveyor belt speed.

2. The intelligent detection method for flow rate of a tea production line according to claim 1 is characterized in that: The process of constructing the prediction model includes: Acquire sample tea image data, process labels, conveyor belt speed labels and actual weight data of the sample tea; the sample tea image data includes sample tea two-dimensional image data and sample tea three-dimensional image data; Obtaining actual volume data of the sample tea leaves based on the three-dimensional image data of the sample tea leaves; Constructing a sample data set based on the process label, the conveyor belt speed label, the actual volume data of the sample tea leaves, and the actual weight data of the sample tea leaves; Constructing an initial prediction model based on the CNN model, the fully connected layer and the optimizer; Inputting the sample tea two-dimensional image data into the initial prediction model to obtain the sample tea predicted process type, conveyor belt predicted speed and sample tea predicted weight data; The initial prediction model is trained based on the sample data set until the output result of the trained initial prediction model meets the set requirements, and the trained initial prediction model is used as the prediction model.

3. The intelligent detection method for flow rate of a tea production line according to claim 2 is characterized in that: Obtaining actual volume data of the sample tea leaves based on the sample tea leaves three-dimensional image data includes: Calibrate the three-dimensional image of the sample tea leaves to obtain calibration image data; Obtaining point cloud data based on the calibration image data; The actual volume data of the sample tea leaves is obtained based on the point cloud data.

4. The intelligent detection method for flow rate of a tea production line according to claim 2 is characterized in that: The actual weight data of the sample tea leaves correspond one to one with the actual volume data of the sample tea leaves; A sample data set is constructed based on the actual volume data of the sample tea leaves and the corresponding actual weight data of the sample tea leaves.

5. The intelligent detection method for flow rate of a tea production line according to claim 2 is characterized in that: Inputting the sample tea two-dimensional image data into the initial prediction model to obtain the sample tea predicted process type, conveyor belt predicted speed and sample tea predicted weight data, including: Normalizing and tensorizing the sample tea two-dimensional image data to obtain processed image data; Based on the processed image data, the initial prediction model is used to obtain the predicted process type of the sample tea leaves, the predicted speed of the conveyor belt and the predicted weight data of the sample tea leaves.

6. The intelligent detection method for flow rate of a tea production line according to claim 5 is characterized in that: Based on the processed image data, the initial prediction model is used to obtain the predicted process type of the sample tea leaves, the predicted speed of the conveyor belt and the predicted weight data of the sample tea leaves, including: Based on the processed image data, a CNN model is used to obtain shared features; Based on the shared features, a fully connected layer is used to obtain the predicted process type of the sample tea leaves and the predicted speed of the conveyor belt; Based on the shared features, the fully connected layer is used to obtain the predicted volume data of the sample tea leaves; Based on the shared features and the predicted volume data of the sample tea leaves, the predicted weight data of the sample tea leaves is obtained using the fully connected layer.

7. The intelligent detection method for flow rate of a tea production line according to claim 2 is characterized in that: The initial prediction model is trained based on the sample data set using an adaptive joint loss function; the adaptive joint loss function includes a mean square error loss function.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the intelligent flow detection method for a tea production line according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the intelligent flow detection method for a tea production line according to any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the intelligent flow detection method for a tea production line according to any one of claims 1 to 7 is implemented.

Images