Method, medium and system for determining plastic discharge cutting time based on infrared light

Abstract

The present invention provides a method, medium and system for determining the time of plastic discharge cutting based on infrared light, which belongs to the field of infrared light measurement and calculation-assisted production technology, and includes: using infrared imaging and image acquisition equipment arranged at the outlet of plastic production equipment to obtain infrared and visible light images of the surface of discharged plastic, and pre-processing. Subsequently, a preset heat conduction model and a pre-trained surface image analysis model are used to predict the internal temperature distribution of the discharged plastic and the shape and size of the contained plastic blocks, respectively, to obtain a temperature distribution matrix and a plastic block distribution matrix. The two matrices are aligned and information is fused to generate a fusion matrix. The fusion matrix is ​​input into a pre-trained plastic block distribution model to calculate the three-dimensional distribution of the plastic block and its internal temperature distribution. Finally, the optimal cutting time is calculated using a cutting empirical equation and output to a cutting tool, which solves the problem of difficulty in determining the optimal cutting time in the prior art.

Description

Technical Field

[0001] The present invention belongs to the field of infrared light measurement and calculation-assisted production technology, and in particular, relates to a method, medium and system for determining plastic discharge cutting timing based on infrared light. Background Art

[0002] The production process of plastic products usually includes key steps such as melt extrusion and cooling molding. The first thing discharged after melting is liquid melted plastic, commonly known as plastic water. This plastic water is recycled and reused. In the production process of plastic discharging and cutting, the plastic water generated and discharged when the machine is turned on directly hits each plastic block, which contains a large number of plastic blocks. Because the plastic blocks are large and difficult to crush, it is difficult to recycle, resulting in a waste of resources. Figure 1 The figure shows a schematic diagram of the cutting knife structure at the outlet of the plastic production equipment. The current method is to rely on manual experience to judge the cutting time, which is often not accurate enough. In addition, since the operator judges the cutting time and then operates the cutting knife motor to cut, cutting lag often occurs.

[0003] That is to say, the existing technology has the problem of difficulty in determining the best cutting time. Summary of the invention

[0004] In view of this, the present invention provides a method, medium and system for determining the plastic discharge cutting time based on infrared light, which can solve the problem in the prior art that it is difficult to determine the best cutting time.

[0005] The present invention is achieved in that:

[0006] A first aspect of the present invention provides a method for determining the plastic discharge cutting time based on infrared light, comprising the following steps:

[0007] S10, acquiring an infrared image and a visible light image of the surface of the discharged plastic collected by an infrared imaging device and an image collection device disposed at the outlet of the plastic production equipment;

[0008] S20, preprocessing the infrared image and the visible light image respectively to obtain a preprocessed infrared image and a preprocessed visible light image;

[0009] S30, predicting the internal temperature distribution of the discharged plastic using a preset heat conduction model according to the preprocessed infrared image, and obtaining a temperature distribution matrix for representing the temperature value of each point of the discharged plastic in a three-dimensional space;

[0010] S40, analyzing the pre-processed visible light image by using a pre-trained surface image analysis model, preliminarily identifying the shape and size of the plastic blocks contained in the discharged plastic, determining the boundaries of the plastic blocks, and obtaining a plastic block distribution matrix for representing the position, shape and size information of each plastic block in the discharged plastic;

[0011] S50, aligning the temperature distribution matrix and the plastic block distribution matrix, and integrating information of the two matrices through a fusion algorithm to obtain a fusion matrix;

[0012] S60, inputting the fusion matrix into a pre-trained plastic block distribution model, and calculating a plastic block matrix for representing the three-dimensional distribution of the plastic block and its internal temperature distribution;

[0013] S70, calculating the cutting time using a cutting empirical equation according to the plastic block matrix and the discharge speed of the discharged plastic, and outputting the calculated cutting time to a cutting tool.

[0014] Specifically, the step S10 specifically includes:

[0015] Step 101, set an infrared imaging device and an image acquisition device at the exit of the plastic production equipment to collect infrared images and visible light images of the surface of the discharged plastic. The infrared imaging device uses the principle of thermal radiation imaging to capture the heat distribution on the surface of the discharged plastic and obtain a high-quality infrared thermal image, which provides a basis for the subsequent analysis of the temperature distribution inside the plastic; the image acquisition device collects visible light images of the discharged plastic to reflect the real appearance characteristics of the plastic, which provides a basis for the subsequent surface shape analysis. Through the synergy of these two imaging devices, rich information including temperature and shape can be obtained, laying the foundation for subsequent multimodal analysis.

[0016] Step 102: Preprocess the infrared image and the visible light image. Filter and denoise the infrared image to eliminate random noise generated during thermal radiation imaging; perform distortion correction on the visible light image to eliminate geometric distortion introduced by the camera lens; perform color correction to ensure that the image color information is consistent with the actual image; and perform image enhancement, such as histogram equalization, to improve image contrast and clarity. After these preprocessing steps, both images have reached sufficient quality and are ready for subsequent multimodal fusion analysis.

[0017] The step S30 specifically includes:

[0018] Step 301: Establish a heat conduction model based on the Fourier heat conduction equation to describe the variation of the temperature inside the plastic with time and space. The heat conduction model includes the Fourier heat conduction equation, the boundary condition equation, the initial condition equation, and the material parameter equation. The Fourier heat conduction equation is used to describe the variation of the temperature inside the plastic with time and space, the boundary condition equation describes the heat exchange process between the plastic surface and the environment, the initial condition equation gives the initial temperature distribution of the plastic, and the material parameter equation describes the variation of the thermophysical properties of the plastic with temperature.

[0019] Step 302: Input the preprocessed infrared image as a boundary condition into the heat conduction model, and obtain a temperature distribution matrix describing the three-dimensional temperature distribution inside the discharged plastic by numerical solution. This provides important temperature field information for subsequent multi-modal fusion analysis.

[0020] The step S40 specifically includes:

[0021] Step 401: Establish a surface image analysis model based on an adversarial neural network, which is used to identify the shape and size of the plastic block contained in the discharged plastic from the visible light image. The model includes a generator network and a discriminator network. The generator network adopts a U-Net structure, the input is the original visible light image, and the output is the contour map of the segmented plastic block; the discriminator network adopts a fully convolutional network structure, the input is the segmentation map, and the output is the authenticity score. Through adversarial training, a model that can accurately segment plastic blocks is finally obtained.

[0022] Step 402: collect a large number of visible light images from the plastic production process, manually annotate the contours of the plastic blocks, and construct a training data set after preprocessing such as data enhancement.

[0023] Step 403: Input the preprocessed visible light image into a pre-trained surface image analysis model to obtain a plastic block distribution matrix describing the three-dimensional distribution of plastic blocks, which provides preliminary shape and size information of the plastic blocks for subsequent multimodal fusion analysis.

[0024] The step S50 specifically includes:

[0025] Step 501, determine the spatial resolution and size of the temperature distribution matrix obtained in step S30 and the plastic block distribution matrix obtained in step S40. If the resolutions are different, the matrix with lower resolution is improved to higher resolution by interpolation method.

[0026] Step 502: Calculate the spatial transformation relationship between the two matrices using an image registration algorithm such as a phase correlation method or a feature point matching method.

[0027] Step 503: Based on the calculated transformation relationship, perform an affine transformation on one of the matrices so that it is aligned with the other matrix in terms of spatial position.

[0028] Step 504: Check the alignment effect and make fine adjustments if necessary.

[0029] Step 505: Integrate the temperature distribution information and the plastic block distribution information into a fusion matrix through a weighted fusion algorithm.

[0030] The purpose of this step is to effectively integrate multimodal information (temperature and shape) to provide complete input data for subsequent three-dimensional plastic block distribution analysis.

[0031] The step S60 specifically includes:

[0032] Step 601, a convolutional neural network is used to construct a plastic block distribution model, including ResNet50 as the backbone network to extract multi-scale features, a feature pyramid network (FPN) to further extract multi-scale features, a region proposal network (RPN) to generate candidate regions, a RoIAlign layer to extract fixed-size features of the candidate regions, a fully connected layer for classification and bounding box regression, and finally non-maximum suppression (NMS) for post-processing.

[0033] Step 602: Input the fusion matrix obtained in step S50 into the plastic block distribution model, and after a series of convolution, pooling, full connection and other operations, finally output a plastic block matrix that accurately describes the three-dimensional distribution, shape, size and internal temperature distribution of the plastic block.

[0034] The step S70 specifically includes:

[0035] Step 701, using the cutting empirical equation to calculate the cutting time, the formula is:

[0036] ;

[0037] in, is the cutting time (s), is the distance between the cutting position and the outlet (m), is the discharge speed (m / s), and are the highest and lowest temperatures of the plastic block (K), is the cutting temperature threshold (K), is the volume of the plastic block (m³), is the maximum cuttable volume (m³), is the coefficient to be determined, is the error term.

[0038] Step 702: output the calculated cutting time to the cutting tool to achieve precise control.

[0039] Through this cutting empirical equation, combined with the temperature and volume information in the plastic block matrix obtained in step S60, the optimal cutting time can be calculated, thereby improving cutting efficiency and quality.

[0040] Among them, the formula of the temperature distribution matrix is ​​expressed as follows:

[0041] ;

[0042] In the formula, Represents the midpoint of three-dimensional space The temperature value at and Represent the number of rows and columns of the matrix respectively.

[0043] The formula of the plastic block distribution matrix is ​​as follows:

[0044] ;

[0045] In the formula, Represents the midpoint of three-dimensional space Is it a plastic block? Indicates that the point belongs to the plastic block, Indicates that the point does not belong to the plastic block.

[0046] The Fourier heat conduction equation is used to describe the change of the temperature inside the plastic with time and space, the input is the spatial coordinates and time, and the output is the temperature;

[0047] The boundary condition equation is used to describe the heat exchange between the plastic surface and the environment, the input is the surface temperature and the ambient temperature, and the output is the heat flux density;

[0048] The initial condition equation is used to describe the initial temperature distribution of the plastic, the input is the spatial coordinates, and the output is the initial temperature;

[0049] The material parameter equation is used to describe the change of the thermophysical properties of plastics with temperature. The input is temperature and the output is thermal conductivity, specific heat capacity and density.

[0050] 1. Fourier heat conduction equation:

[0051] ;

[0052] In the formula, is the density (kg / m³); is the specific heat capacity (J / (kg·K)); is temperature (K); is time(s); is the thermal conductivity (W / (m·K)); is the internal heat source term (W / m³).

[0053] 2. Boundary condition equation:

[0054] ;

[0055] In the formula, is the surface normal vector; is the convection heat transfer coefficient (W / (m²·K)); is the surface temperature (K); is the ambient temperature (K); is the surface emissivity; is the Stefan-Boltzmann constant (W / (m²·K 4 )).

[0056] 3. Initial condition equation:

[0057] ;

[0058] In the formula, is the initial temperature distribution function.

[0059] 4. Material parameter equation:

[0060] ;

[0061] ;

[0062] ;

[0063] In the formula, are unknown coefficients obtained by fitting experimental data.

[0064] Parameter acquisition method:

[0065] 1. The method is obtained by experimental measurement, and the steps include: (1) preparing a standard sample; (2) using a thermal analyzer to measure the specific heat capacity at different temperatures; (3) using a thermal conductivity meter to measure the thermal conductivity at different temperatures; and (4) using a densitometer to measure the density at different temperatures.

[0066] 2. The results are obtained by combining experiments with numerical simulations, and the steps include: (1) building an experimental platform to measure the change of plastic surface temperature over time; (2) establishing a numerical model to obtain the convective heat transfer coefficient by solving an inverse problem.

[0067] 3. The method is obtained by measuring with an infrared thermometer, and the steps include: (1) preparing a plastic sample; (2) measuring the surface temperature of the sample using an infrared thermometer; and (3) calculating the emissivity by comparing with the actual temperature.

[0068] 4. The method is obtained by fitting historical data, and the steps include: (1) collecting historical cutting data; (2) fitting parameters of the cutting empirical equation using the least squares method; and (3) verifying the accuracy of the fitting result.

[0069] Further, the temperature distribution matrix and the plastic block distribution matrix are aligned, specifically:

[0070] First, the spatial resolution and size of the two matrices are determined;

[0071] If the resolutions are different, the lower resolution matrix is ​​interpolated to the higher resolution;

[0072] Use image registration algorithm to calculate the spatial transformation relationship between two matrices;

[0073] According to the calculated transformation relationship, an affine transformation is performed on one of the matrices to align it with the other matrix in spatial position.

[0074] Furthermore, the surface image analysis model adopts an adversarial neural network, specifically comprising:

[0075] Generator network: uses the U-Net structure, with the input being the original visible light image and the output being the contour image of the segmented plastic block;

[0076] Discriminator network: uses a fully convolutional network structure, with the input being a segmentation map and the output being a authenticity score;

[0077] The loss functions include adversarial loss, pixel-level L1 loss, and perceptual loss;

[0078] During the training process, the generator network strives to generate a real segmentation map, and the discriminator strives to distinguish between the real and generated segmentation maps;

[0079] By alternately training the generator and the discriminator, the surface image analysis model for accurately segmenting plastic blocks is finally obtained.

[0080] Furthermore, the step of establishing a training data set for the surface image analysis model specifically includes:

[0081] Collecting visible light images of a large number of plastic production processes;

[0082] Manually annotate the outlines of plastic blocks in the image;

[0083] Data enhancement: including random cropping, rotation, flipping, and adjusting brightness and contrast;

[0084] The dataset is divided into training set, validation set and test set in a ratio of 8:1:1;

[0085] Normalize the image so that the pixel values ​​are distributed between 0 and 1;

[0086] The images and corresponding annotation information are packaged into tfrecord format for quick reading during model training.

[0087] Furthermore, the training step of the surface image analysis model specifically includes:

[0088] Initialize the generator and discriminator network parameters;

[0089] Load the preprocessed training dataset;

[0090] Training the discriminator: input the real segmentation map and the segmentation map generated by the generator, and minimize the cross entropy loss between the two;

[0091] Train the generator: fix the discriminator parameters, minimize the loss of the generated image being judged as fake, as well as the L1 distance and perceptual distance to the real segmentation map;

[0092] Alternate the steps of training the discriminator and training the generator until the model converges.

[0093] Furthermore, the plastic block distribution model adopts a convolutional neural network model, uses ResNet50 as the backbone network, and also includes:

[0094] Feature Pyramid Network (FPN), used to extract multi-scale features;

[0095] Region Proposal Network (RPN), used to generate candidate regions;

[0096] RoIAlign layer, used to extract fixed-size feature maps of candidate regions;

[0097] Fully connected layers for classification and bounding box regression;

[0098] Non-maximum suppression (NMS) post-processing is used to remove overlapping detection boxes.

[0099] Furthermore, the training step of the plastic block distribution model specifically includes:

[0100] Using steps S10-S50 to obtain fusion matrices corresponding to multiple groups of discharged plastics;

[0101] Using a temperature probe sequence arranged in the outlet of the plastic production equipment to detect the temperature of the multiple groups of discharged plastics, and obtain temperature distribution data inside the discharged plastics;

[0102] Segmenting and detecting the multiple groups of discharged plastics to obtain three-dimensional distribution data of plastic blocks inside the discharged plastics;

[0103] The temperature distribution data and the three-dimensional distribution data of the plastic block are combined into a training plastic block matrix;

[0104] Establishing a training data set, including fusion matrices corresponding to the multiple groups of discharged plastics and their corresponding training plastic block matrices;

[0105] The convolutional neural network model is trained using the training data set to obtain the plastic block distribution model.

[0106] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned method for determining the plastic discharge cutting time based on infrared light.

[0107] The third aspect of the present invention provides a cutting point determination system for plastic production discharge, which includes the above-mentioned computer-readable storage medium and a host computer, an infrared imaging device and an image acquisition device, wherein the infrared imaging device and the image acquisition device are both communicatively connected to the host computer, and the host computer is connected to the control system of the cutting tool via a data cable; the readable storage medium is arranged in the host computer, and the host computer is used to execute the program instructions stored in the readable storage medium.

[0108] Compared with the prior art, the method, medium and system for determining the time of plastic material discharge cutting based on infrared light provided by the present invention have the following beneficial effects:

[0109] 1. Integrate temperature and shape information to improve the accuracy of cutting timing prediction

[0110] Traditional visual analysis algorithms only focus on the geometric features of the plastic surface, ignoring the important impact of internal temperature distribution on cutting timing. The present invention obtains the internal temperature field information of the plastic through infrared thermal imaging technology, and integrates it with the surface shape information obtained by visible light imaging for analysis, which can more comprehensively and accurately predict the optimal cutting time and greatly improve the cutting quality.

[0111] 2. Realize intelligence and automation to improve production efficiency

[0112] Manual experience in judging the cutting timing is easily affected by subjective factors and is difficult to achieve high automation. The present invention uses computer vision and intelligent algorithms to automatically monitor and analyze the temperature distribution and shape changes of plastics, realize intelligent prediction and control of cutting timing, and greatly improve production efficiency and product quality consistency.

[0113] In summary, the present invention solves the problem in the prior art that it is difficult to determine the best cutting time. BRIEF DESCRIPTION OF THE DRAWINGS

[0114] Figure 1 It is a schematic diagram of the structure of a cutting knife at the outlet of a plastic production equipment in the background technology of the present invention;

[0115] Figure 2 is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0116] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0117] like Figure 2 FIG. 1 is a flow chart of a method for determining the time of plastic material discharge cutting based on infrared light provided by the present invention. The method comprises the following steps:

[0118] S10, acquiring an infrared image and a visible light image of the surface of the discharged plastic collected by an infrared imaging device and an image collection device disposed at the outlet of the plastic production equipment;

[0119] S20, preprocessing the infrared image and the visible light image respectively to obtain a preprocessed infrared image and a preprocessed visible light image;

[0120] S30, predicting the internal temperature distribution of the discharged plastic using a preset heat conduction model based on the preprocessed infrared image, and obtaining a temperature distribution matrix for representing the temperature value of each point of the discharged plastic in the three-dimensional space;

[0121] S40, analyzing the preprocessed visible light image by using a pre-trained surface image analysis model, preliminarily identifying the shape and size of the plastic blocks contained in the discharged plastic, determining the boundaries of the plastic blocks, and obtaining a plastic block distribution matrix for representing the position, shape and size information of each plastic block in the discharged plastic;

[0122] S50, aligning the temperature distribution matrix and the plastic block distribution matrix, and integrating the information of the two matrices through a fusion algorithm to obtain a fusion matrix;

[0123] S60, inputting the fusion matrix into a pre-trained plastic block distribution model, and calculating a plastic block matrix for representing the three-dimensional distribution of the plastic block and its internal temperature distribution;

[0124] S70, according to the plastic block matrix and the discharge speed of the discharged plastic, using the cutting empirical equation, calculate the cutting time, and output it to the cutting tool.

[0125] The specific implementation methods of the above steps are described in detail below:

[0126] First, step S10 is to obtain the infrared image and visible light image of the surface of the discharged plastic collected by the infrared imaging device and the image acquisition device set at the outlet of the plastic production equipment. The purpose of this step is to obtain the temperature distribution information and appearance shape information of the discharged plastic. Specifically:

[0127] 1) Infrared imaging equipment uses the principle of thermal radiation imaging to capture the heat distribution discharged from the plastic surface and obtain high-quality infrared thermal images. This helps to analyze the temperature distribution inside the plastic in the subsequent analysis.

[0128] 2) The image acquisition device collects visible light images of the discharged plastic, reflecting the true appearance characteristics of the plastic and providing a basis for subsequent surface shape analysis.

[0129] Through the synergistic effect of these two imaging devices, rich information including temperature and shape can be obtained, laying the foundation for subsequent multimodal analysis.

[0130] Next is step S20, preprocessing the acquired infrared image and visible light image. The purpose of preprocessing is to remove noise and correct distortion so that the image reflects the real state of the plastic more clearly and accurately. Specifically, it includes:

[0131] 1) Filter and denoise the infrared image to eliminate the random noise generated during the thermal radiation imaging process.

[0132] 2) Perform distortion correction on visible light images to eliminate the geometric distortion introduced by the camera lens.

[0133] 3) Color correction to ensure that the image color information is consistent with the actual situation.

[0134] 4) Image enhancement, such as histogram equalization, to improve image contrast and clarity.

[0135] After these preprocessing steps, both images reached sufficient quality and were ready for subsequent multimodal fusion analysis.

[0136] Then, step S30 is performed to predict the internal temperature distribution of the discharged plastic using a preset heat conduction model based on the pre-processed infrared image. The key to this step is to establish an accurate heat conduction model that can accurately simulate the temperature field changes inside the plastic. Specifically, it includes:

[0137] 1) Establish a mathematical model based on Fourier heat conduction equation to describe the variation of internal temperature of plastic with time and space.

[0138] 2) Set reasonable boundary condition equations to describe the heat exchange process between the plastic surface and the environment.

[0139] 3) Determine the initial condition equation and give the initial temperature distribution of the plastic.

[0140] 4) Establish material parameter equations to describe how the thermophysical properties of plastics such as thermal conductivity, specific heat capacity and density change with temperature.

[0141] By solving these coupled partial differential equations, a complete heat conduction model can be obtained. By inputting the infrared image as the boundary condition into the model, the three-dimensional temperature distribution matrix inside the discharged plastic can be predicted. This provides important temperature field information for subsequent multimodal fusion analysis.

[0142] Next is step S40, where the visible light image is analyzed by a pre-trained surface image analysis model to preliminarily identify the shape and size of the plastic blocks contained in the discharged plastic. This model is trained using an adversarial neural network, including:

[0143] 1) Generator network: Using the U-Net structure, the input is the original visible light image, and the output is the segmented plastic block contour map.

[0144] 2) Discriminator network: uses a fully convolutional network structure, with the input being a segmentation map and the output being a authenticity score.

[0145] 3) During training, the generator strives to generate realistic segmentation maps, while the discriminator strives to distinguish between true and false. The two achieve optimal results through adversarial training.

[0146] 4) Through this adversarial training method, we finally get a model that can accurately segment plastic blocks.

[0147] The training data set of this model is constructed by collecting a large number of visible light images from the plastic production process, manually annotating the contours of plastic blocks, and performing preprocessing such as data enhancement.

[0148] By applying this pre-trained surface image analysis model, the preliminary shape and size information of the plastic block can be obtained from the visible light image, laying the foundation for subsequent multimodal fusion analysis.

[0149] Step S50 is to align and fuse the temperature distribution matrix obtained in step S30 and the plastic block distribution matrix obtained in step S40. Specifically, it includes:

[0150] 1) First, determine the spatial resolution and size of the two matrices. If the resolutions are different, the matrix with lower resolution is increased to higher resolution through interpolation.

[0151] 2) Use image registration algorithms such as phase correlation method or feature point matching method to calculate the spatial transformation relationship between the two matrices.

[0152] 3) Based on the calculated transformation relationship, perform an affine transformation on one of the matrices so that it is aligned with the other matrix in terms of spatial position.

[0153] 4) Check the alignment and make fine adjustments if necessary.

[0154] 5) Finally, the temperature distribution information and plastic block distribution information are integrated into a fusion matrix through the weighted fusion algorithm.

[0155] The purpose of this step is to effectively integrate multimodal information (temperature and shape) to provide complete input data for subsequent three-dimensional plastic block distribution analysis.

[0156] Next is step S60, where the fusion matrix is ​​input into the pre-trained plastic block distribution model to calculate a plastic block matrix for representing the three-dimensional distribution of the plastic block and its internal temperature distribution. This model uses a convolutional neural network, including:

[0157] 1) ResNet50 is used as the backbone network to extract multi-scale features.

[0158] 2) Feature Pyramid Network (FPN), used to further extract multi-scale features.

[0159] 3) Region Proposal Network (RPN), generates candidate regions.

[0160] 4) RoIAlign layer, extracts fixed-size features of the candidate region.

[0161] 5) Fully connected layer, performs classification and bounding box regression.

[0162] 6) Non-maximum suppression (NMS) post-processing to remove overlapping detection boxes.

[0163] The input of the model is the fusion matrix obtained in step S50, which contains temperature and preliminary shape information. After being processed by this model, a plastic block matrix that accurately describes the three-dimensional distribution, shape, size and internal temperature distribution of the plastic block can be obtained.

[0164] Finally, step S70 is to calculate the cutting time using the cutting empirical equation according to the plastic block matrix obtained in step S60 and the discharge speed of the discharged plastic, and output it to the cutting tool. The cutting empirical equation is:

[0165] ;

[0166] in, is the cutting time (s), is the distance between the cutting position and the outlet (m), is the discharge speed (m / s), and are the highest and lowest temperatures of the plastic block (K), is the cutting temperature threshold (K), is the volume of the plastic block (m³), is the maximum cuttable volume (m³), is the coefficient to be determined, is the error term.

[0167] Through this empirical cutting equation, combined with the temperature and volume information in the plastic block matrix, the optimal cutting time can be calculated, thereby accurately controlling the cutting tool and improving cutting efficiency and quality.

[0168] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned method for determining the plastic discharge cutting time based on infrared light.

[0169] The third aspect of the present invention provides a cutting point determination system for plastic production and discharging, which includes the above-mentioned computer-readable storage medium and a host computer, an infrared imaging device and an image acquisition device, wherein the infrared imaging device and the image acquisition device are both connected to the host computer in communication, and the host computer is connected to the control system of the cutting tool via a data cable. Generally speaking, the host computer adopts an industrial computer, and the infrared thermal imaging device adopts the FLIR A655sc model, which has an image resolution of 640×480, a temperature measurement accuracy of ±2°C, and a frame rate of up to 50 frames / second. The device can reflect the distribution of the internal temperature field by measuring the thermal radiation flux on the plastic surface; the visible light imaging device uses the Sony Alpha 7R IV model, with a resolution of up to 61,000 pixels, which can capture real plastic appearance images. The optical axes of the two devices are installed in parallel to ensure that the temperature and shape information of the same area can be obtained at the same time.

[0170] Specifically, the principle of the present invention is:

[0171] 1. Multimodal imaging and preprocessing

[0172] First, the method uses infrared imaging equipment and visible light imaging equipment to obtain the temperature distribution information and geometric appearance information of the plastic surface respectively. The infrared imaging equipment uses the principle of thermal radiation imaging to capture the heat distribution on the plastic surface, thereby reflecting the changes in the internal temperature field. The visible light imaging equipment collects the real appearance characteristics of the plastic surface, such as shape, size, etc.

[0173] The acquired infrared images and visible light images need to be preprocessed, including denoising, distortion correction, color correction, etc., to ensure that the image quality meets the requirements of subsequent analysis.

[0174] 2. Heat conduction model and temperature distribution prediction

[0175] In order to accurately extract the temperature distribution information inside the plastic from the infrared image, the present invention establishes a mathematical model based on the Fourier heat conduction equation. The model describes the variation of the temperature inside the plastic with time and space, covering the heat conduction equation, boundary condition equation, initial condition equation and material parameter equation. By solving these coupled partial differential equations, a complete heat conduction model can be obtained, and the infrared image is used as the boundary condition input to predict the three-dimensional temperature distribution matrix inside the plastic.

[0176] 3. Surface image analysis and plastic block distribution identification

[0177] At the same time, the present invention adopts a surface image analysis model trained by an adversarial neural network to identify and segment the geometric information such as the shape and size of the plastic block from the visible light image. The model includes a generator network and a discriminator network, and through adversarial training, a model that can accurately segment the plastic block is finally obtained.

[0178] 4. Multimodal information fusion and 3D reconstruction of plastic blocks

[0179] The temperature distribution matrix and the plastic block distribution matrix obtained in step 2 and step 3 are aligned and fused to form a fusion matrix containing temperature and shape information. When fusion, first ensure that the two matrices are consistent in spatial resolution and scale, then use the image registration algorithm to calculate the transformation relationship, and perform affine transformation alignment on one of the matrices. Finally, the two types of information are combined through a weighted fusion algorithm.

[0180] The fusion matrix is ​​taken as input to a pre-trained convolutional neural network model, which can further extract multi-scale features, accurately identify and locate plastic blocks, estimate their three-dimensional spatial distribution, shape and size, and internal temperature distribution, and output a complete plastic block matrix.

[0181] 5. Cutting timing prediction and control

[0182] Finally, the present invention uses the cutting empirical equation to calculate the best cutting time based on the plastic block matrix obtained in step 4 and combined with parameters such as the discharge speed. The cutting empirical equation takes into account factors such as the temperature distribution and size of the plastic block and can more accurately predict the cutting time. The calculation results are fed back to the cutting tool to achieve intelligent cutting control.

[0183] Through the above five key steps, the multimodal fusion method of the present invention can make full use of the internal temperature distribution and surface shape information obtained by infrared thermal imaging and visible light imaging, and after a series of technical processing such as heat conduction modeling and neural network analysis, it can finally accurately determine the optimal plastic cutting time, thereby greatly improving the cutting quality and promoting the intelligence and automation of the plastic production process.

[0184] A specific embodiment 1 of the method of the present invention is provided below. The steps of this embodiment 1 are described in detail as follows: First, step S10 is to obtain an infrared image and a visible light image of the surface of the discharged plastic collected by an infrared imaging device and an image collection device set at the outlet of the plastic production equipment. The purpose of this step is to obtain the temperature distribution information and appearance shape information of the discharged plastic. Specifically:

[0185] Infrared imaging equipment uses the principle of thermal radiation imaging to measure the thermal radiation flux on the plastic surface. To obtain temperature distribution information. According to the Stefan-Boltzmann law, the thermal radiation flux The surface temperature of the object The relationship is:

[0186] ;

[0187] In the formula, is the emissivity of the plastic surface, is the Stefan-Boltzmann constant, is the ambient temperature. By analyzing the distribution of the temperature on the surface of the discharged plastic, we can get the temperature distribution image.

[0188] The image acquisition device collects visible light images of the discharged plastic to reflect the true appearance characteristics of the plastic. The visible light image contains rich information such as geometric shape and material characteristics, providing a basis for subsequent surface shape analysis.

[0189] By obtaining temperature distribution information through infrared imaging and appearance shape information through visible light imaging, the physical state of the discharged plastic can be comprehensively characterized, laying the foundation for subsequent multimodal analysis.

[0190] Next is step S20, preprocessing the acquired infrared image and visible light image. The purpose of preprocessing is to remove noise and correct distortion so that the image reflects the real state of the plastic more clearly and accurately. Specifically, it includes:

[0191] 1) To filter and denoise the infrared image, a Gaussian filter can be used, and its filter kernel is:

[0192] ;

[0193] In the formula, is the standard deviation of the filter, which controls the strength of the filter. The filter can effectively eliminate the random noise generated in the thermal radiation imaging process.

[0194] 2) To correct the distortion of visible light images, a camera calibration method can be used. First, determine the camera's intrinsic parameter matrix and the distortion parameter vector , and then use the following formula to perform distortion correction:

[0195] ;

[0196] in, is the distortion parameter, , is the original image coordinate, are the corrected coordinates.

[0197] 3) Perform color correction to ensure that the image color information is consistent with the actual color. You can use the color management system to establish a transformation matrix from the original RGB value to the standard sRGB value. , and then apply the transformation:

[0198] ;

[0199] 4) Image enhancement, such as histogram equalization, improves image contrast and clarity. Histogram equalization can be achieved using the following formula:

[0200] ;

[0201] in, is the original gray value, is the gray value after equalization, is the gray level, is the image size, Gray value The number of pixels.

[0202] After these preprocessing steps, both images reached sufficient quality and were ready for subsequent multimodal fusion analysis.

[0203] Then, step S30 is performed to predict the internal temperature distribution of the discharged plastic using a preset heat conduction model based on the pre-processed infrared image. The key to this step is to establish an accurate heat conduction model that can accurately simulate the temperature field changes inside the plastic. Specifically, it includes:

[0204] 1) Establish a mathematical model based on Fourier heat conduction equation to describe the internal temperature of the plastic Over time and space The changing rules:

[0205] ;

[0206] In the formula, is the density, is the specific heat capacity, is the thermal conductivity, is the internal heat source term.

[0207] 2) Set reasonable boundary condition equations to describe the heat exchange process between the plastic surface and the environment:

[0208] ;

[0209] In the formula, is the surface normal vector, is the convective heat transfer coefficient, is the surface emissivity, is the surface temperature, is the ambient temperature.

[0210] 3) Determine the initial condition equation and give the initial temperature distribution of the plastic:

[0211] ;

[0212] 4) Establish material parameter equations to describe the changes in the thermophysical properties of plastics with temperature:

[0213] ;

[0214]

[0215] ;

[0216] By solving these coupled partial differential equations, a complete heat conduction model can be obtained. By inputting the infrared image as the boundary condition into the model, the three-dimensional temperature distribution matrix of the discharged plastic can be predicted. ,in Representing a spatial point This provides important temperature field information for subsequent multimodal fusion analysis.

[0217] Next is step S40, where the visible light image is analyzed by a pre-trained surface image analysis model to preliminarily identify the shape and size of the plastic blocks contained in the discharged plastic. This model is trained using an adversarial neural network, including a generator network and a discriminator network:

[0218] The generator network adopts the U-Net structure, and the input is the original visible light image , the output is the contour of the segmented plastic block The U-Net network can effectively extract multi-scale feature information from the original image through encoding-decoding and generate fine segmentation results.

[0219] The discriminator network adopts a fully convolutional network structure, and the input is the segmentation map output by the generator. , the output is the authenticity score The discriminator network attempts to distinguish whether the generated segmentation map is different from the real segmentation map annotated by humans. Figure 1 To.

[0220] During training, the generator Trying to generate realistic segmentation maps To minimize the following loss function:

[0221] ;

[0222] in, is the real segmentation map, is the L1 norm, For perceived loss.

[0223] Discriminator We try to minimize the following loss function to distinguish true and false segmentation maps:

[0224] ;

[0225] Through adversarial training, a model that can accurately segment plastic blocks is finally obtained. Applying this pre-trained surface image analysis model, the preliminary shape and size information of plastic blocks can be obtained from visible light images, laying the foundation for subsequent multimodal fusion analysis.

[0226] Step S50 is to convert the temperature distribution matrix obtained in step S30 into and the plastic block distribution matrix obtained in step S40 Perform alignment and fusion. Specifically include:

[0227] 1) First determine the spatial resolution and size of the two matrices. If the resolutions are different, the lower resolution matrix is ​​interpolated to a higher resolution. For example, for the temperature distribution matrix , cubic spline interpolation can be used:

[0228] ;

[0229] in, are the basis functions in the x, y, and z directions.

[0230] 2) Use image registration algorithms such as phase correlation method or feature point matching method to calculate the spatial transformation relationship between the two matrices The phase correlation method can be implemented through Fourier transform:

[0231] ;

[0232] In the formula, is the Fourier transform, For conjugation.

[0233] 3) According to the calculated transformation relationship , for one of the matrices (such as ) to perform an affine transformation on it so that it is consistent with another matrix (such as ) are aligned in space.

[0234] 4) Check the alignment and make fine adjustments if necessary.

[0235] 5) Finally, the temperature distribution information and the plastic block distribution information are integrated into a fusion matrix through the weighted fusion algorithm. For example:

[0236] ;

[0237] in, is the weight coefficient, which can be adjusted according to the importance of the two types of information.

[0238] The purpose of this step is to effectively integrate multimodal information (temperature and shape) to provide complete input data for subsequent three-dimensional plastic block distribution analysis.

[0239] Next is step S60, the fusion matrix Input into the pre-trained plastic block distribution model to calculate the plastic block matrix used to represent the three-dimensional distribution of the plastic block and its internal temperature distribution This model uses a convolutional neural network, including:

[0240] 1) ResNet50 is used as the backbone network to extract multi-scale features. By introducing residual connections, the ResNet network can effectively avoid the gradient vanishing problem and extract richer feature representations.

[0241] 2) Feature Pyramid Network (FPN), which is used to further extract multi-scale features. FPN can fully utilize the advantages of features of different scales through bottom-up and top-down feature fusion.

[0242] 3) Region Proposal Network (RPN), generates candidate regions. RPN predicts possible regions containing objects by sliding windows.

[0243] 4) RoIAlign layer, extracts fixed-size features of the candidate region. RoIAlign can avoid quantization errors and better preserve feature information.

[0244] 5) Fully connected layer, which performs classification and bounding box regression. The classification outputs whether each candidate area is a plastic block, and the regression outputs the precise bounding box position and size.

[0245] 6) Non-maximum suppression (NMS) post-processing to remove overlapping detection boxes.

[0246] The input of this model is the fusion matrix obtained in step S50 , including temperature and preliminary shape information. After processing by this model, a plastic block matrix can be obtained that accurately describes the three-dimensional distribution, shape, size and internal temperature distribution of the plastic block. ,in Representing a spatial point Is it a plastic block?

[0247] Finally, step S70 is performed to obtain the plastic block matrix obtained in step S60. and the discharge speed of the discharged plastic , using the cutting empirical equation to calculate the cutting time , and output to the cutting tool. The cutting empirical equation is:

[0248] ;

[0249] in, is the distance between the cutting position and the exit, and are the highest and lowest temperatures of the plastic block, is the cutting temperature threshold, is the volume of the plastic block, is the maximum cuttable volume, is the coefficient to be determined, is the error term.

[0250] Through this cutting empirical equation, combined with the plastic block matrix The temperature and volume information in the image can be used to calculate the optimal cutting time. , thereby accurately controlling the cutting tool and improving cutting efficiency and quality.

[0251] In summary, in this embodiment 1, the temperature and shape information obtained by infrared imaging and visible light imaging are fully utilized, and through the processing and fusion of a series of mathematical modeling and neural network algorithms, the cutting time is finally accurately determined, providing strong support for the intelligence and automation of plastic production equipment.

[0252] In order to better understand and implement the present invention, the following is an example 2 of a specific application scenario of the present invention: A certain enterprise is mainly engaged in the manufacture of plastic extrusion products, including plastic profiles, pipes, etc. In the production process, cutting control has always been a key link, affecting the quality and production efficiency of the product. In the past, the enterprise mainly relied on experienced operators to observe the plastic extrusion conditions and determine the cutting time based on subjective judgment. However, this method is difficult to achieve high automation and is easily affected by human factors, resulting in large plastic blocks that are difficult to crush when the machine is turned on, making it difficult to recycle and causing waste of resources.

[0253] Construction of multimodal imaging system:

[0254] At the exit of the plastic production equipment, the company installed a multimodal imaging system, including infrared thermal imaging equipment and visible light imaging equipment.

[0255] The infrared thermal imaging device uses the FLIRA655sc model, which has an image resolution of 640×480, a temperature measurement accuracy of ±2℃, and a frame rate of up to 50 frames per second. The device can reflect the distribution of the internal temperature field by measuring the thermal radiation flux on the plastic surface.

[0256] The visible light imaging device is a Sony Alpha7RIV model with a resolution of up to 61,000 pixels, which can capture real plastic appearance images. The optical axes of the two devices are installed in parallel to ensure that the temperature and shape information of the same area can be obtained at the same time.

[0257] In order to ensure image quality, the company is also equipped with corresponding image preprocessing equipment, including Gaussian filters, distortion correction modules, etc., to perform noise suppression, geometric correction and other operations on the collected infrared images and visible light images.

[0258] Establishment of heat conduction model and prediction of temperature distribution:

[0259] In order to accurately extract the temperature distribution information inside the plastic from the infrared image, the company established a mathematical model based on the Fourier heat conduction equation. Specifically, it includes:

[0260] Heat conduction equation:

[0261] ;

[0262] in, is the density of the plastic, is the specific heat capacity, is the thermal conductivity, is the internal heat source term.

[0263] Boundary condition equation:

[0264] ;

[0265] here, is the convective heat transfer coefficient, is the emissivity, is the ambient temperature.

[0266] Initial condition equation:

[0267] ;

[0268] Material parameter equation:

[0269] ;

[0270] ;

[0271] ;

[0272] By solving the above coupled equations by finite element method, the temperature matrix describing the three-dimensional temperature distribution inside the plastic can be obtained: Take a production process as an example. Some of the data are shown in Table 1:

[0273] Table 1 Plastic internal temperature distribution matrix ;

[0274]

[0275] As can be seen from Table 1, through the calculation of the heat conduction model, the temperature value of the plastic at each point in the three-dimensional space is obtained, which lays the foundation for the subsequent multimodal fusion analysis.

[0276] Surface image analysis and plastic block distribution identification:

[0277] At the same time, the company has also established a surface image analysis model based on adversarial neural networks to identify and segment geometric information such as the shape and size of plastic blocks from visible light images.

[0278] The generator network of this model adopts U-Net structure, and the input is visible light image , the output is the contour of the segmented plastic block The discriminator network uses a fully convolutional network, and the input is the output of the generator , the output is the authenticity score .

[0279] During training, the generator Trying to generate realistic segmentation maps , to minimize the following loss function:

[0280] ;

[0281] in, is the real segmentation map annotated manually, is the L1 norm, For perceived loss.

[0282] Discriminator We try to minimize the following loss function to distinguish true and false segmentation maps:

[0283] ;

[0284] By alternately training the generator and the discriminator, we eventually obtain a model that can accurately segment plastic blocks.

[0285] Using this model, the preliminary shape and size information of the plastic block is extracted from the visible light images collected during a production process, as shown in Table 2:

[0286] Table 2 Plastic block distribution matrix ;

[0287]

[0288] In Table 2, Indicates the point Belong to plastic blocks, It indicates that it does not belong to the category. It can be seen that the model successfully extracts the preliminary distribution information of plastic blocks from the visible light image.

[0289] Multimodal information fusion and 3D reconstruction of plastic blocks:

[0290] Next, the temperature distribution matrix obtained in step 2 and step 3 is and plastic block distribution matrix Perform alignment fusion to form a fusion matrix containing temperature and shape information .

[0291] First, ensure that the spatial resolution and scale of the two matrices are consistent. and The resolution is different, Perform cubic spline interpolation to raise it to the same value as Same resolution.

[0292] Then, the spatial transformation relationship between the two matrices is calculated using the phase correlation method. :

[0293] ;

[0294] according to right Perform an affine transformation to make it Align in spatial position. Finally, the weighted fusion algorithm is used to obtain the fusion matrix :

[0295] ;

[0296] Fusion Matrix It contains two key information: temperature and shape, providing complete input data for subsequent three-dimensional plastic block analysis.

[0297] Next, The input is fed into a pre-trained convolutional neural network model, which can further extract multi-scale features, accurately identify and locate plastic blocks, and estimate their three-dimensional spatial distribution, shape and size, and internal temperature distribution. The output of the model is a complete plastic block matrix , as shown in Table 3:

[0298] Table 3 Plastic block matrix ;

[0299]

[0300] As can be seen from Table 3, the model not only accurately identifies the three-dimensional distribution of plastic blocks, but also estimates the detailed information such as temperature, length, width and height of each plastic block. This provides a sufficient basis for the subsequent prediction of cutting timing.

[0301] Cutting timing prediction and control:

[0302] Finally, according to the plastic block information in Table 3, combined with parameters such as discharge speed, the optimal cutting time is calculated using the cutting empirical equation:

[0303] ;

[0304] in, is the distance between the cutting position and the exit, is the discharge speed, and are the highest and lowest temperatures of the plastic block in Table 3, is the volume of the plastic block.

[0305] According to the above formula, the optimal cutting time is This result is fed back to the cutting tool to achieve intelligent cutting control.

[0306] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for determining the plastic discharge cutting time based on infrared light, characterized in that: The following steps are involved: S10, acquiring an infrared image and a visible light image of the surface of the discharged plastic collected by an infrared imaging device and an image collection device disposed at the outlet of the plastic production equipment; S20, preprocessing the infrared image and the visible light image respectively to obtain a preprocessed infrared image and a preprocessed visible light image; S30, predicting the internal temperature distribution of the discharged plastic using a preset heat conduction model according to the preprocessed infrared image, and obtaining a temperature distribution matrix for representing the temperature value of each point of the discharged plastic in a three-dimensional space; S40, analyzing the pre-processed visible light image by using a pre-trained surface image analysis model, preliminarily identifying the shape and size of the plastic blocks contained in the discharged plastic, determining the boundaries of the plastic blocks, and obtaining a plastic block distribution matrix for representing the position, shape and size information of each plastic block in the discharged plastic; S50, aligning the temperature distribution matrix and the plastic block distribution matrix, and integrating information of the two matrices through a fusion algorithm to obtain a fusion matrix; S60, inputting the fusion matrix into a pre-trained plastic block distribution model, and calculating a plastic block matrix for representing the three-dimensional distribution of the plastic block and its internal temperature distribution; S70, calculating the cutting time using a cutting empirical equation according to the plastic block matrix and the discharge speed of the discharged plastic, and outputting the calculated cutting time to a cutting tool.

2. The method for determining the plastic discharge cutting time based on infrared light according to claim 1, characterized in that: The heat conduction model includes Fourier heat conduction equation, boundary condition equation, initial condition equation and material parameter equation; The Fourier heat conduction equation is used to describe the change of temperature inside the plastic with time and space. The input is the spatial coordinate and time, and the output is the temperature. The boundary condition equation is used to describe the heat exchange between the plastic surface and the environment, the input is the surface temperature and the ambient temperature, and the output is the heat flux density; The initial condition equation is used to describe the initial temperature distribution of the plastic, the input is the spatial coordinates, and the output is the initial temperature; The material parameter equation is used to describe the change of the thermophysical properties of plastics with temperature. The input is temperature and the output is thermal conductivity, specific heat capacity and density.

3. The method for determining the plastic discharge cutting time based on infrared light according to claim 2, characterized in that: The temperature distribution matrix and the plastic block distribution matrix are aligned, specifically: First, the spatial resolution and size of the two matrices are determined; If the resolutions are different, the lower resolution matrix is ​​interpolated to the higher resolution; Use image registration algorithm to calculate the spatial transformation relationship between two matrices; According to the calculated transformation relationship, an affine transformation is performed on one of the matrices to align it with the other matrix in spatial position.

4. The method for determining the plastic discharge cutting time based on infrared light according to claim 3, characterized in that: The surface image analysis model adopts an adversarial neural network, specifically comprising: Generator network: uses the U-Net structure, with the input being the original visible light image and the output being the contour image of the segmented plastic block; Discriminator network: uses a fully convolutional network structure, with the input being a segmentation map and the output being a authenticity score; The loss functions include adversarial loss, pixel-level L1 loss, and perceptual loss; During the training process, the generator network strives to generate a real segmentation map, and the discriminator strives to distinguish between the real and generated segmentation maps; By alternately training the generator and the discriminator, the surface image analysis model for accurately segmenting plastic blocks is finally obtained.

5. The method for determining the plastic discharge cutting time based on infrared light according to claim 4, characterized in that: The step of establishing the training data set of the surface image analysis model specifically includes: Collecting visible light images of a large number of plastic production processes; Manually annotate the outlines of plastic blocks in the image; Data enhancement: including random cropping, rotation, flipping, and adjusting brightness and contrast; The dataset is divided into training set, validation set and test set in a ratio of 8:1:1; Normalize the image so that the pixel values ​​are distributed between 0 and 1; The images and corresponding annotation information are packaged into tfrecord format for quick reading during model training.

6. The method for determining the plastic discharge cutting time based on infrared light according to claim 5, characterized in that: The training step of the surface image analysis model specifically includes: Initialize the generator and discriminator network parameters; Load the preprocessed training dataset; Training the discriminator: input the real segmentation map and the segmentation map generated by the generator, and minimize the cross entropy loss between the two; Train the generator: fix the discriminator parameters, minimize the loss of the generated image being judged as fake, as well as the L1 distance and perceptual distance to the real segmentation map; Alternate the steps of training the discriminator and training the generator until the model converges.

7. The method for determining the plastic discharge cutting time based on infrared light according to claim 6, characterized in that: The plastic block distribution model adopts a convolutional neural network model, uses ResNet50 as the backbone network, and also includes: Feature pyramid network, used to extract multi-scale features; The region proposal network is used to generate candidate regions; RoIAlign layer, used to extract fixed-size feature maps of candidate regions; Fully connected layers for classification and bounding box regression; Non-maximum suppression post-processing is used to remove overlapping detection boxes.

8. The method for determining the plastic discharge cutting time based on infrared light according to claim 7, characterized in that: The training steps of the plastic block distribution model specifically include: Using steps S10-S50 to obtain fusion matrices corresponding to multiple groups of discharged plastics; Using a temperature probe sequence arranged in the outlet of the plastic production equipment to detect the temperature of the multiple groups of discharged plastics, and obtain temperature distribution data inside the discharged plastics; Segmenting and detecting the multiple groups of discharged plastics to obtain three-dimensional distribution data of plastic blocks inside the discharged plastics; The temperature distribution data and the three-dimensional distribution data of the plastic block are combined into a training plastic block matrix; Establishing a training data set, including fusion matrices corresponding to the multiple groups of discharged plastics and their corresponding training plastic block matrices; The convolutional neural network model is trained using the training data set to obtain the plastic block distribution model.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed in a computer, the method for determining the plastic discharge cutting time based on infrared light according to any one of claims 1 to 8 is used.

10. A system for determining cutting points of plastic production and discharging, characterized in that: It comprises the computer-readable storage medium as claimed in claim 9, and a host computer, an infrared imaging device and an image acquisition device, wherein the infrared imaging device and the image acquisition device are both communicatively connected to the host computer, and the host computer is connected to the control system of the cutting tool via a data cable; the readable storage medium is arranged in the host computer, and the host computer is used to execute the program instructions stored in the readable storage medium.

Images