Powder pneumatic conveying image processing method and equipment and storage medium

Abstract

The invention relates to a powder pneumatic conveying image processing method and device and a storage medium, and is applied to the field of image processing, and the method comprises the steps: obtaining an image of a material accumulation contour of a feed port and corresponding feeding text information, and setting the image of the material accumulation contour of the feed port as an original contour image; preprocessing the original contour image by adopting an adaptive contrast enhancement fusion color correction algorithm, and generating a target contour image; adding a preset number of Gaussian noise to the target contour image to generate a noise contour image; and inputting the noise contour image and the corresponding feeding text information into the trained contour enhancement neural network model, and outputting a clear contour image after Bayesian denoising. The method has the technical effects that a material accumulation image with a clear outline is obtained, and the dynamic change process of the material accumulation outline at the feed port is seen.

Description

Technical Field

[0001] The present application relates to the technical field of image data processing, and in particular to a method, device and storage medium for processing images of powder pneumatic conveying. Background Art

[0002] In the pharmaceutical, food, chemical, new energy and other industries, ton bags, small bags of powder or granular materials are usually fed manually or by a robotic arm or other conveyor to the feeding port during the feeding process. The manual feeding method is generally inefficient. Feeding materials to the feeding port by a robotic arm or other conveyor can solve the problem of low efficiency of manual feeding. However, when the robotic arm feeds, it may be difficult to find the blockage in time, which may further affect the efficiency and cause material waste. There are many conventional parameters for monitoring the working status of the feed port, including the feed port level meter, flow rate, air pressure value in the pipe, etc. The working status of the feed port is usually monitored by video images; however, when using video images to monitor the working status of the feed port, since the powder will generate dust during operation, the video image captured by the camera will contain powder dust, resulting in blurry and unclear video images captured by the camera.

[0003] At present, there are some image data processing methods that process the image of the feed port to obtain clear video images; however, after processing using the current image processing methods, it is still difficult to clearly see the material accumulation outline at the feed port through the image, which makes it difficult to judge whether there is a serious accumulation trend at the discharge port based on the material accumulation outline at the feed port, thereby causing blockage and affecting the feeding efficiency. Summary of the invention

[0004] In order to help solve the problem that it is difficult to clearly see the material accumulation outline at the feed port through the image, which makes it difficult to judge whether there is a serious accumulation trend at the feed port based on the material accumulation outline at the feed port, causing blockage and affecting the feeding efficiency, the present application provides a powder pneumatic conveying image processing method, equipment and storage medium.

[0005] In a first aspect, the present application provides a method for processing an image of powder pneumatic conveying, which adopts the following technical solution: the method comprises:

[0006] Acquire an image of the material accumulation contour of the feed inlet and corresponding feeding text information, and set the image of the material accumulation contour of the feed inlet as an original contour image;

[0007] Adopting an adaptive contrast enhancement fusion color correction algorithm to preprocess the original contour image and generate a target contour image;

[0008] Adding a preset number of Gaussian noises to the target contour image to generate a noise contour image;

[0009] The noise contour image and the corresponding feeding text information are input into the trained contour enhancement neural network model, and a clear contour image is output after Bayesian denoising.

[0010] In a specific implementation scheme, the original contour image is a color image in RGB format, and the method of preprocessing the original contour image using an adaptive contrast enhancement fusion color correction algorithm to generate a target contour image includes:

[0011] The original contour image in RGB format is converted into the original contour image in HSI format, wherein the original contour image in HSI format includes an H channel, an S channel, and an I channel. The conversion method includes:

[0012]

[0013] H=360°-H(if B>G)

[0014]

[0015] Among them, R, G, B represent the data of R channel, G channel and B channel of the original contour image in RGB format, respectively, and H, S, I represent the data of H channel, S channel and I channel of the original contour image in HSI format after conversion, respectively;

[0016] The data of the H channel, the S channel and the I channel are respectively normalized to generate a normalized H array, a normalized S array and a normalized I array. The calculation method of the data normalization of the H channel, the S channel and the I channel includes:

[0017] S norm =S,

[0018] Among them, H norm represents the normalized H array, S norm represents the normalized S array, I norm represents the normalized I array, I max Indicates the maximum value in the I channel data;

[0019] Adopting an adaptive contrast enhancement algorithm, a hue correction algorithm, and a saturation enhancement algorithm to process the normalized I array, the normalized H array, and the normalized S array, respectively, and generating an enhanced I array, a corrected H array, and an enhanced S array, respectively;

[0020] The corrected H array, the enhanced S array and the enhanced I array are merged to generate a corrected HSI image;

[0021] The corrected HSI image is converted into an image in RGB format, and the target contour image is generated. The conversion method includes:

[0022] (R, G, B) = ((R′+m)×255, (G′+m)×255, (B′+m)×255),

[0023] in,

[0024] C=Inorm′·S′

[0025] X=C·(1-|(H′×6)mod2-1|)

[0026]

[0027] Among them, (R, G, B) represents the RGB value of the target contour image in RGB form after conversion, R', G', B', C, X and m are intermediate calculation variables, Inorm' represents the enhanced I array, S' represents the enhanced S array, and H' represents the corrected H array.

[0028] In a specific implementation scheme, the adopting of the adaptive contrast enhancement algorithm, the hue correction algorithm, and the saturation enhancement algorithm to process the normalized I array, the normalized H array, and the normalized S array, respectively, and generating an enhanced I array, a corrected H array, and an enhanced S array, respectively, comprises:

[0029] Dividing the image corresponding to the normalized I array into a plurality of I channel grid blocks according to a preset size;

[0030] Calculate the minimum pixel value, maximum pixel value, grid block mean value and grid block standard deviation value of each of the I channel grid blocks;

[0031] The normalized I array is adjusted according to the minimum pixel value, the maximum pixel value, the grid block mean value, and the grid block standard deviation value, and the adjusted I array is set as the enhanced I array, and the adjustment method includes:

[0032]

[0033] Among them, Inorm′ n Indicates the adjusted nth pixel value, Inorm n Represents the nth pixel value of the image corresponding to the normalized I array, I min represents the minimum pixel value, I max represents the maximum pixel value, and L represents the maximum range of pixel values;

[0034] Calculate the H channel local mean and H channel local standard deviation of the preset area of ​​the normalized H array, and the H channel global mean and H channel global standard deviation of the global normalized H array;

[0035] The normalized H array is corrected according to the H channel local mean, the H channel local standard deviation, the H channel global mean, and the H channel global standard deviation, and the corrected H array is set as the corrected H array, and the correction method includes:

[0036]

[0037] Among them, H′ n Indicates the nth value of the calibration H array, H norm，n represents the nth value of the normalized H array, μ H represents the local mean of the H channel, σ H Represents the local standard deviation of the H channel, Represents the global standard deviation of the H channel, represents the global mean of H channel;

[0038] Calculating the S channel local mean and the S channel local standard deviation of a preset area of ​​the normalized S array;

[0039] The normalized S array is adjusted according to the local mean value of the S channel and the local standard deviation value of the S channel, and the adjusted S array is set as the enhanced S array. The adjustment method includes:

[0040]

[0041] Among them, α and β represent adaptive adjustment parameters, S′ n represents the nth value of the enhanced S array, S norm，n represents the nth value of the normalized S array, μ S represents the local mean of the S channel, σ S Indicates the local standard deviation of the S channel.

[0042] In a specific implementation scheme, the training method of the contour enhancement neural network model includes:

[0043] Acquire a material accumulation clear outline image set and feeding text information corresponding to the material accumulation clear outline image set, wherein the material accumulation clear outline image set is a collection of material accumulation clear outline images at a material inlet;

[0044] Using a histogram equalization algorithm to pre-process each image in the material accumulation clear contour image set, and generate a balanced contour image set;

[0045] Adding a preset number of Gaussian noises to each image in the equalized contour image set, and generating a fuzzy contour image set;

[0046] Using a contour enhancement neural network model to fuse each image in the fuzzy contour image set with the corresponding feed text information, and generate a fused image set;

[0047] De-noising each image in the fused image set using a Bayesian denoising algorithm to generate a de-noised contour image set;

[0048] Calculate the image data difference between each image in the denoised contour image set and the corresponding image in the equalized contour image set;

[0049] If the image data difference is within a preset range, the contour recognition neural network model training is completed.

[0050] In a specific implementation scheme, the use of a contour enhancement neural network model to fuse each image in the fuzzy contour image set with the corresponding feed text information and generate a fused image set includes:

[0051] Encoding the feeding text information and generating feeding text encoding data;

[0052] Merging the data of each image in the fuzzy contour image set with the corresponding input text encoding data to generate image-text fusion data;

[0053] The image-text fusion data is converted into a fused image by using a contour enhancement neural network model, and the fused image set is generated.

[0054] In a specific implementation manner, after outputting the clear contour image, the method further comprises:

[0055] Acquire the feeding text information corresponding to the original contour image, fuse the clear contour image with the feeding text information corresponding to the original contour image, and generate edge prediction original data;

[0056] The edge prediction raw data is input into the trained edge prediction neural network model, and the predicted edge contour image is output.

[0057] In a specific implementation scheme, the training method of the edge prediction neural network model includes:

[0058] Constructing an edge prediction neural network model, wherein the edge prediction neural network model includes six convolutional layers and five pooling layers;

[0059] Inputting the edge prediction raw data into the edge prediction neural network model for feature extraction, and outputting a predicted edge image;

[0060] A loss function is constructed according to the material type, and the edge prediction neural network model is optimized according to the loss function to generate a trained edge prediction neural network model.

[0061] In a specific implementation scheme, inputting the edge prediction raw data into the edge prediction neural network model for feature extraction, and outputting the predicted edge image comprises:

[0062] The data output by the seventh convolutional layer is upsampled by 4 times to generate the upsampled data of the seventh convolutional layer, and the data output by the fourth pooling layer is upsampled by 2 times to generate the upsampled data of the fourth pooling layer;

[0063] Fusing the upsampled data of the seventh convolutional layer, the upsampled data of the fourth pooling layer, and the data output by the third pooling layer to generate convolutional pooling fusion data;

[0064] The convolution pooling fusion data is upsampled 8 times, and the predicted edge image is output.

[0065] In a second aspect, the present application provides a powder pneumatic conveying image processing device, which adopts the following technical solution: the device comprises:

[0066] An image acquisition module is used to acquire an image of the material accumulation contour at the feed inlet and corresponding feeding text information, and set the image of the material accumulation contour at the feed inlet as an original contour image;

[0067] An equalization processing module, used for preprocessing the original contour image by using a histogram equalization algorithm, and generating a target contour image;

[0068] A noise adding module, used for adding a preset number of Gaussian noises to the target contour image to generate a noise contour image;

[0069] The image output module is used to input the noise contour image and the corresponding feeding text information into the trained contour enhancement neural network model, and output a clear contour image after Bayesian denoising.

[0070] In a third aspect, the present application provides a computer device, which adopts the following technical solution: it includes a memory and a processor, and the memory stores a computer program that can be loaded by the processor and execute any of the above-mentioned powder pneumatic conveying image processing methods.

[0071] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution: a computer program that can be loaded by a processor and execute any of the above-mentioned powder pneumatic conveying image processing methods is stored.

[0072] In summary, this application has the following beneficial technical effects:

[0073] The original contour image is first preprocessed and then input into the trained neural network model for calculation, and a clear contour image is output. The obtained clear contour image has a clear contour and a clear edge. Based on the clear contour image, the accumulation trend of the feed port can be judged, so as to avoid blockage as much as possible and thus improve the feeding efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 is a flow chart of the powder pneumatic conveying image processing method in an embodiment of the present application;

[0075] Figure 2 It is a schematic diagram of the framework of the contour enhancement neural network model training in the embodiment of the present application;

[0076] Figure 3 Schematic diagram of the convolutional network structure of the contour enhancement neural network model in the embodiment of the present application;

[0077] Figure 4 This is a schematic diagram of feature extraction of the edge prediction neural network model in an embodiment of the present application;

[0078] Figure 5 is a schematic diagram of an image processing device for powder pneumatic conveying in an embodiment of the present application;

[0079] Figure 6 It is a schematic diagram used to embody a computer device in an embodiment of the present application.

[0080] Reference numerals: 501, image acquisition module; 502, equalization processing module; 503, noise adding module; 504, image output module. DETAILED DESCRIPTION

[0081] The following combination Figure 1-Figure 6 This application is described in further detail.

[0082] An embodiment of the present application discloses a powder pneumatic conveying image processing method, which is applied to a powder pneumatic conveying image processing system. The image processing method can be used to process an image of a material accumulation contour at a feed port in a feeding process in powder pneumatic conveying, so that the image contour is clearer and the contour edge is clearer, thereby avoiding the problem of material blockage in the feeding process of a robotic arm as much as possible and improving the feeding efficiency of the robotic arm.

[0083] In the pharmaceutical, food, chemical, new energy and other industries, ton bags, small bags of powder or granular materials are usually fed manually or by a robotic arm or other conveyor to the feeding port during the feeding process. The manual feeding method is generally inefficient. Feeding materials to the feeding port by a robotic arm or other conveyor can solve the problem of low efficiency of manual feeding. However, when the robotic arm feeds, it may be difficult to find the blockage in time, which may further affect the efficiency and cause material waste. There are many conventional parameters for monitoring the working status of the feed port, including the feed port level meter, flow rate, air pressure value in the pipe, etc. The working status of the feed port is usually monitored by video images; however, when using video images to monitor the working status of the feed port, since the powder will generate dust during operation, the video image captured by the camera will contain powder dust, resulting in blurry and unclear video images captured by the camera.

[0084] At present, there are some image data processing methods that process the image of the feed port to obtain a clear video image; however, after processing using the current image processing method, it is still difficult to clearly see the material accumulation contour at the feed port through the image, and it is also difficult to see the dynamic change process of the material accumulation contour at the feed port, which makes it difficult to judge whether there is a serious accumulation trend at the feed port based on the material accumulation contour at the feed port, which in turn causes blockage and affects the feeding efficiency. It may also affect material pollution and cause economic losses. In order to help improve the clarity of the contour image, the dynamic change process of the accumulation contour can be seen from the image, and the feeding efficiency can be improved. This application provides a powder pneumatic conveying image processing method.

[0085] Reference Figure 1 , the method comprises the following steps:

[0086] S10, obtaining an image of the material accumulation contour at the feed inlet and corresponding feeding text information, and setting the image of the material accumulation contour at the feed inlet as an original contour image.

[0087] Specifically, under normal circumstances, users will use a camera to capture video images to monitor the working status of the feed port. The video images captured by the camera are processed to obtain each frame of the image, and the frame images of the obtained video images are input into the image processing system. The system obtains the image of the material accumulation contour at the feed port, and sets the image of the material accumulation contour at the feed port as the original contour image. The corresponding feeding text information refers to the type of material corresponding to the image, and obtains external factors such as the temperature, altitude, air pressure, humidity, lighting conditions, and particulate matter (PM2.5) values ​​under the current conditions of the image. For example, the text information obtained by the system can be expressed as "Flour 36℃57cm 1009hpa 36%20%122μg / m 3"; Inputting the text information and the contour image together into the neural network model can increase the weight of the text information, so that it can affect the contour image to obtain a contour image with higher definition.

[0088] S20, preprocessing the original contour image using an adaptive contrast enhancement fusion color correction algorithm, and generating a target contour image.

[0089] Specifically, an adaptive contrast enhancement fusion color correction algorithm is used to preprocess the original contour image. The adaptive contrast enhancement fusion color correction algorithm includes an adaptive contrast enhancement algorithm and a color correction fusion algorithm, so that the brightness and color distribution of the image can be adjusted more intelligently.

[0090] S30, adding a preset number of Gaussian noises to the target contour image to generate a noise contour image.

[0091] Specifically, a preset number of Gaussian noises are added to the target image obtained after preprocessing, wherein the method of adding Gaussian noises is cumulative noise addition, which can also be understood as adding Gaussian noises to the target image for the first time, and obtaining an image with Gaussian noises added once, adding Gaussian noises to the image with Gaussian noises added once for the second time, and obtaining an image with Gaussian noises added twice, and so on, to obtain an image with Gaussian noises added a preset number of times; specifically, for example, assuming that the target image obtained after preprocessing is X0, the number of times of adding Gaussian noises is set to k, generally speaking, the range of k can be set to [5, 15], then the noise image obtained after adding Gaussian noises for the tth time can be expressed as:

[0092]

[0093] Among them, X0 represents the target image before adding noise; X t represents the noisy image after adding Gaussian noise for the tth time; Z t Represents the mean of the Gaussian random variables added from 1 to t. All the added Gaussian random variables obey the Gaussian distribution of N(0,1) and have a mean Z t It also obeys the Gaussian distribution of N(0,1); a t represents the weight of the Gaussian noise in the noise image obtained by adding (t-1) times Gaussian noise, a t The calculation method can be expressed as:

[0094] a t =1-β t

[0095] Among them, β t is a pre-set value, β t The range is [0.0001, 0.02], βt As t increases, it increases proportionally, with an initial value of 0.0001 and increasing to a maximum value of 0.02; after k times of superposition, the final noise contour image is obtained.

[0096] S40, inputting the noise contour image and the corresponding feeding text information into the trained contour enhancement neural network model, performing Bayesian denoising and outputting a clear contour image.

[0097] Specifically, the obtained noise contour image and the corresponding feeding text information are input into the trained contour enhancement neural network model. The trained neural network model is used to influence the contour image information according to the text information, and the weight influence of the text information is increased, which helps to obtain a contour image with higher clarity. Since the input is a noisy contour image, it is necessary to denoise the image output by the model to obtain a clear contour image. Specifically, denoising can be performed using Bayesian denoising. For example, assuming that the noisy image output by the neural network is Y k After k times of denoising, the image obtained is Y0, where k times of denoising corresponds to the number of times of adding noise; the image obtained after the (t-1)th denoising is Y t-1 It can be expressed as:

[0098]

[0099] Among them, a t It represents the weight of the noise in the noisy image obtained after (t-1) times of denoising, which is different from the a in the added Gaussian noise. t The meaning of the expression is consistent; θ t Represents a variable parameter, with an initial value of 1, which can be continuously adjusted according to the difference between the training results and the actual target; σ t represents the adjustment coefficient, σ t The calculation method can be expressed as:

[0100]

[0101] Finally, after k times of denoising, a clear contour image is obtained.

[0102] In the present application, the original contour image is preprocessed before being input into the trained neural network model for calculation, and a clear contour image is output. The obtained clear contour image is an image with clear contours and clear edges. Based on the clear contour image, the accumulation trend of the feed port can be determined, so as to avoid material blockage as much as possible, thereby improving the feeding efficiency. Among them, the preprocessing adopts the histogram equalization algorithm, which is a preprocessing method for dust in a relatively dim environment, which can improve the image quality, so that the preprocessed image can show more details hidden by the external environment.

[0103] In one embodiment, the original contour image is a color image in RGB format, and the steps of preprocessing the original contour image using an adaptive contrast enhancement fusion color correction algorithm and generating a target contour image can be specifically performed as follows:

[0104] First, the original contour image in RGB format is converted to the original contour image in HSI format. The original contour image in HSI format includes H channel, S channel and I channel, which represent hue, saturation and intensity respectively. The conversion calculation formula is as follows:

[0105]

[0106] H=360°-H(if B>G)

[0107]

[0108] Among them, R, G, B represent the data of R channel, G channel and B channel of the original contour image in RGB format, respectively, and H, S, I represent the data of H channel, S channel and I channel of the original contour image in HSI format after conversion, respectively;

[0109] Subsequently, the data of the H channel, S channel, and I channel are normalized to generate a normalized H array, a normalized S array, and a normalized I array. The normalization calculation formula can be expressed as:

[0110] S norm =S,

[0111] Among them, H norm represents the normalized H array, S norm represents the normalized S array, I norm Represents the normalized I array, I max Indicates the maximum value in the I channel data;

[0112] Afterwards, the adaptive contrast enhancement algorithm, the hue correction algorithm, and the saturation enhancement algorithm are used to process the normalized I array, the normalized H array, and the normalized S array, respectively, and generate an enhanced I array, a corrected H array, and an enhanced S array, respectively; specifically, the normalized I array I norm An adaptive contrast enhancement algorithm is applied, and the normalized H array and the normalized S array are processed using a color correction fusion algorithm.

[0113] First, the image corresponding to the normalized I array is divided into several I channel grid blocks according to a preset size, and the size of each area can be adjusted according to the image resolution and characteristics. After that, local statistical calculations are performed to calculate the minimum pixel value, maximum pixel value, grid block mean and grid block standard deviation value of each I channel grid block. Finally, adaptive adjustment is performed to adjust the normalized I array according to the minimum pixel value, maximum pixel value, grid block mean and grid block standard deviation value, and the adjusted I array is set as the enhanced I array. The adjustment method includes:

[0114]

[0115] Among them, Inorm′ n Indicates the adjusted nth pixel value, Inorm n Represents the nth pixel value of the image corresponding to the normalized I array, I min Indicates the minimum pixel value, I max represents the maximum pixel value, and L represents the maximum range of pixel values.

[0116] Afterwards, the data of the H channel is processed. First, local statistical calculation is performed to calculate the local mean value and local standard deviation value of the H channel of the preset area of ​​the normalized H array, as well as the global mean value and global standard deviation value of the H channel of the normalized H array. Afterwards, a tone equalization process is performed, and the normalized H array is corrected according to the local mean value, the local standard deviation value, the global mean value and the global standard deviation value of the H channel, and the corrected H array is set as the corrected H array. The correction method includes:

[0117]

[0118] Among them, H′ n Indicates the nth value of the H array, H norm,n Represents the normalized nth value of the H array, μ H represents the local mean of the H channel, σ H Represents the local standard deviation of the H channel, Represents the global standard deviation of the H channel, Represents the global mean of the H channel. The global mean and standard deviation of the H channel are used to maintain the consistency of the overall tone.

[0119] After that, the data of the S channel is processed. First, local statistical calculation is performed to calculate the local mean value of the S channel and the local standard deviation value of the S channel in the preset area of ​​the normalized S array. After that, saturation enhancement processing is performed, and the normalized S array is adjusted according to the local mean value of the S channel and the local standard deviation value of the S channel, and the adjusted S array is set as the enhanced S array. The adjustment method includes:

[0120]

[0121] Among them, α and β represent adaptive adjustment parameters, which are dynamically adjusted according to the saturation distribution of the local area to ensure that the saturation is enhanced within a reasonable range. n Indicates enhancing the nth value of the S array, S norm,n Represents the normalized nth value of the S array, μ S represents the local mean of the S channel, σ S Indicates the local standard deviation of the S channel.

[0122] After processing the data of each channel, the adjusted and corrected H array, enhanced S array and enhanced I array are fused to generate a corrected HSI image. The corrected HSI image can be expressed as:

[0123]

[0124] Among them, HSInew represents the corrected HSI image, H' represents the corrected H array, S' represents the enhanced S array, and Inorm' represents the enhanced I array.

[0125] Finally, the corrected HSI image is converted into an RGB image and a target contour image is generated. The conversion methods include:

[0126] (R, G, B) = ((R′+m)×255, (G′+m)×255, (B′+m)×255),

[0127] in,

[0128] C=Inorm′·S′

[0129] X=C·(1-|(H′×6)mod2-1|)

[0130]

[0131] Among them, (R, G, B) represents the RGB value of the target contour image in RGB form after conversion, R', G', B', C, X and m are intermediate calculation variables, Inorm' represents the enhanced I array, S' represents the enhanced S array, and H' represents the corrected H array.

[0132] The adaptive contrast enhancement fusion color correction algorithm is used to pre-process the original image, which can more intelligently adjust the brightness and color distribution of the image, mainly including: 1) Local adaptive enhancement: Adaptively adjust the contrast and color according to the brightness and color distribution of the local area of ​​the image to avoid excessive enhancement or detail loss caused by global adjustment; 2) Hue balance and saturation enhancement: Make the color distribution more uniform through hue balance, and enhance the saturation to improve the color vividness of the image while avoiding color distortion; 3) Keep the color natural: By introducing global and local statistical parameters, ensure that the color-adjusted image remains natural and realistic.

[0133] The traditional preprocessing method is to directly perform histogram equalization on the RGB image. However, the image of the material accumulation contour at the feed port has the problems of relatively single target, more details, and insufficient light. Therefore, the traditional histogram containing RGB color channels is not required for image processing. Based on the sensitivity of light, after converting the RGB image into an HSI image, the adaptive contrast enhancement fusion color correction algorithm is used for processing, which can more effectively optimize the brightness and color distribution, obtain a well-distributed histogram, thereby improving the image quality and showing more details hidden by the external environment.

[0134] In one embodiment, referring to Figure 2 , the steps of training the contour enhancement neural network model can be specifically performed as follows:

[0135] First, obtain a set of clear images of material accumulation contours and the feeding text information corresponding to the set of clear images of material accumulation contours. The set of clear images of material accumulation contours is a set of clear images of material accumulation contours at the material inlet. The image set input for model training is a set of frame images split from video images. The corresponding feeding text information refers to the type of material corresponding to the image. The external factors such as temperature, altitude, air pressure, humidity, lighting conditions, and particulate matter (PM2.5) values ​​under the current conditions of the image are obtained. It should be clear that the image input for the training model is an image of clear material accumulation contours at the material inlet. The size of all input images is 512*512, and the corresponding text information input exists as a correction condition for model training.

[0136] Afterwards, the histogram equalization algorithm is used to preprocess each image in the material accumulation clear contour image set, and a balanced contour image set is generated; the preprocessing method for obtaining the balanced contour image set is consistent with the preprocessing method for obtaining the target contour image, first converting the RGB image into the HSI image, normalizing the data in the three channels of H, S, and I in the HSI image, and then using the histogram equalization processing algorithm, and then converting the HSI image set into the RGB image, and finally obtaining the balanced contour image set. Since the powder pneumatic conveying environment is relatively closed, preprocessing the image can significantly reduce the difficulty of training, and the size of each image in the equalized image set is also 512*512. After preprocessing, a preset number of Gaussian noises are added to each image in the equalized contour image set, and a fuzzy contour image set is generated; specifically, the method of adding Gaussian noise is consistent with the method of obtaining the noisy contour image in step S30, and a fuzzy contour image is obtained after adding noise to the image several times cumulatively. The size of the processed image is 512*512*3, where 3 represents the three channels of R, G, and B. The set of all fuzzy contour images is the fuzzy contour image set.

[0137] After adding Gaussian noise, the contour enhancement neural network model is used to fuse each image in the fuzzy contour image set with the corresponding feeding text information, and a fused image set is generated; specifically, the feeding text information is first encoded, and the feeding text encoding data is generated. The text information format obtained by the system is "flour 36℃57cm 1009hpa 36%20%122μg / m 3 ", each parameter is separated by a space, and there is no need to perform word segmentation on the text input data, which reduces the workload and increases the efficiency of feature extraction; since the input text information contains Chinese information, the characters need to be encoded first, and utf-8 is used for encoding. Then, a 3*1 convolution kernel is used to adjust the size of the text information, and the final size of the text information is 512*1; then, the data of each image in the fuzzy contour image set is fused with the corresponding input text encoding data to generate image-text fusion data, and the size of the fusion data is 512*512*4, where 4 represents the three channels of R, G, and B and text data. Finally, the contour enhancement neural network model is used to convert the image-text fusion data into a fused image, and a fused image set is generated; the purpose of using neural network fusion data is to increase the weight of text information so that the weight can affect the image information. For details, refer to Figure 3, the convolution kernel network can be designed as an eight-layer convolution structure, divided into four layers of downsampling and four layers of upsampling. A nonlinear transformation is added between each convolution layer to prevent feature redundancy between convolution layers; secondly, for example, the data of the 256*256*32 layer is transmitted to the 256*256*16 layer, the purpose is to transfer the features extracted in the front to the back convolution layer as a reference to improve the accuracy of the model. After the output of the neural network model, the size of each image is 512*512*3.

[0138] After obtaining the fused image set, each image in the fused image set is denoised using the Bayesian denoising algorithm to generate a denoised contour image set; the specific Bayesian denoising method is consistent with the Bayesian denoising method used in step S40, and after adding several times of Gaussian noise, the corresponding number of denoising is performed, and the specific denoising method is not repeated here. Finally, the image data difference between each image in the denoised contour image set and the corresponding image in the equalized contour image set is calculated; for example, assuming that the denoised contour image set is Y, the equalized contour image set is X, and the number of images in the image set is N, then the image data difference calculation method can be expressed as:

[0139]

[0140] Among them, i represents the i-th image; if the image data difference is within the preset range, the contour recognition neural network model training is completed, and the preset value is the value of loss1. In the embodiment of the present application, the preset value is set to 20. The size of the preset value can be set according to the actual image processing situation, and there is no limitation here.

[0141] It should be noted that in the embodiments of the present application, an image set is used as a training set for example. An image set can be understood as an image set obtained through a video image. In actual applications, model training can be performed by collecting multiple video images. Multiple video images can increase data samples and improve the accuracy of the trained model.

[0142] In the present application scheme, the model is trained through training and images to obtain a contour enhancement neural network model with higher accuracy. In practical applications, the trained neural network model can be used to obtain an image with clear edges and no artifacts.

[0143] In one embodiment, considering that the material falls at a fast speed and the edge changes are highly real-time, the neural network for image processing will take up a certain amount of computing time and computing resources, which may take a long time. Therefore, after outputting a clear contour image, the following steps may be performed:

[0144] Obtain the feeding text information corresponding to the original contour image, fuse the clear contour image with the feeding text information corresponding to the original contour image, and generate edge prediction raw data. Specifically, the text information includes the type of material corresponding to the image, and obtains the temperature, altitude, air pressure, humidity, lighting conditions, particulate matter (PM2.5) values ​​and other external factors of the current image. For example, the text information obtained by the system can be expressed as "flour 36℃57cm 1009hpa 36%20%122μg / m 3 ", because the text information contains Chinese, it is also necessary to encode the text information first. The size of the obtained clear contour image is 512*512*3, and the size of the edge prediction raw data obtained after combining with the text data is 512*512*4; then, the edge prediction raw data is input into the trained edge prediction neural network model, and the predicted edge contour image is output.

[0145] In the present application, since the material falls at a fast speed and the edge changes are highly real-time, the image processing neural network will take up a certain amount of computing time and computing resources, which may take a long time. In order to achieve effective real-time monitoring, a time-saving and highly accurate edge prediction neural network model is designed and trained to predict the edge contour; in addition, the addition of text data can greatly reduce the data required for reasoning.

[0146] In one embodiment, the training of the edge prediction neural network model may be specifically performed as follows:

[0147] First, an edge prediction neural network model is constructed. The edge prediction neural network model includes six convolutional layers and five pooling layers. The connection order of the convolutional layers and the pooling layers in the constructed edge prediction neural network model can be: the first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer, the third convolutional layer, the third pooling layer, the fourth convolutional layer, the fourth pooling layer, the fifth convolutional layer, the fifth pooling layer, the sixth convolutional layer and the seventh convolutional layer; the size of the input data is 512*512*4, and the input data is convolved to obtain the first convolutional layer. The size of the first convolutional layer is 51 2*512*4, then perform pooling operation, the data size is 256*256*16, then perform convolution operation again, get the second convolution layer, the size is 256*256*16, after pooling becomes 128*128*64, continue to perform convolution and pooling operations, the data needed to be input into the prediction network is 64*64*256, continue the above operation, get 32*32*1024, and so on, the fifth pooling layer becomes 16*16*4096, then after two more convolution operations, the data obtained after the two convolution operations are 16*16*4096.

[0148] Afterwards, the edge prediction raw data is input into the edge prediction neural network model for feature extraction, and the predicted edge image is output; for details, refer to Figure 4 , upsample the data output by the seventh convolution layer by 4 times to generate the upsampled data of the seventh convolution layer, upsample the data output by the fourth pooling layer by 2 times to generate the upsampled data of the fourth pooling layer; fuse the upsampled data of the seventh convolution layer, the upsampled data of the fourth pooling layer and the output data of the third pooling layer to generate the convolution-pooling fusion data; upsample the convolution-pooling fusion data by 8 times to output the predicted edge image.

[0149] Finally, a loss function is constructed according to the material type, and the edge prediction neural network model is optimized according to the loss function to generate a trained edge prediction neural network model; the constructed loss function can be expressed as:

[0150]

[0151] Where N represents (the type of material input + 1). For example, if there is only a single material in the input this time, the value of N is 2; i It represents the label of material type i, where the positive class is 1, which can be understood as this type, and the negative class is 0, which can be understood as not this type; It represents the predicted value; in most cases, the situation where the negative material type is far greater than the positive material type will not occur, so the above loss function method is adopted in the embodiment of the present application.

[0152] It should be noted that in an embodiment of the present application, the set of clear contour images in the edge prediction original data set can directly adopt the set of clear contour images obtained in the above-mentioned contour enhancement neural network model, and the model can be trained and optimized through multiple image data samples in the data set to improve the accuracy of model prediction.

[0153] In the present application scheme, the model is trained through training and images to obtain a more accurate edge prediction neural network model. In practical applications, the trained neural network model can be used to obtain an edge image with a higher prediction accuracy.

[0154] Figure 1 FIG. 1 is a flow chart of a method for processing powder pneumatic conveying images in one embodiment. It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but the steps are not necessarily executed in the order indicated by the arrows; unless otherwise specified in this document, there is no strict order restriction for the execution of the steps, and the steps may be executed in other orders; and Figure 1At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0155] Based on the above method, an embodiment of the present application also discloses a powder pneumatic conveying image processing device.

[0156] Reference Figure 5 , the device includes the following modules:

[0157] An image acquisition module 501 is used to acquire an image of the material accumulation contour at the feed inlet and corresponding feeding text information, and set the image of the material accumulation contour at the feed inlet as an original contour image;

[0158] The equalization processing module 502 is used to pre-process the original contour image using an adaptive contrast enhancement fusion color correction algorithm and generate a target contour image;

[0159] The noise adding module 503 is used to add a preset number of Gaussian noises to the target contour image to generate a noise contour image;

[0160] The image output module 504 is used to input the noise contour image and the corresponding feeding text information into the trained contour enhancement neural network model, and output a clear contour image after performing Bayesian denoising.

[0161] In one embodiment, the equalization processing module 502 is specifically used to convert the original contour image in RGB format into the original contour image in HSI format, where the original contour image in HSI format includes an H channel, an S channel, and an I channel. The conversion method includes:

[0162]

[0163] H=360°-H(if B>G)

[0164]

[0165] Among them, R, G, B represent the data of R channel, G channel and B channel of the original contour image in RGB format, respectively, and H, S, I represent the data of H channel, S channel and I channel of the original contour image in HSI format after conversion, respectively;

[0166] The data of the H channel, S channel, and I channel are normalized respectively to generate a normalized H array, a normalized S array, and a normalized I array. The calculation method of the data normalization of the H channel, S channel, and I channel includes:

[0167] S norm =S,

[0168] Among them, H norm represents the normalized H array, S norm represents the normalized S array, I norm Represents the normalized I array, I max Indicates the maximum value in the I channel data;

[0169] Adopting adaptive contrast enhancement algorithm, hue correction algorithm and saturation enhancement algorithm to process normalized I array, normalized H array and normalized S array respectively, and generate enhanced I array, corrected H array and enhanced S array respectively; fuse the corrected H array, enhanced S array and enhanced I array to generate corrected HSI image; convert the corrected HSI image into RGB image and generate target contour image, the conversion method includes:

[0170] (R, G, B) = ((R′+m)×255, (G′+m)×255, (B′+m)×255),

[0171] in,

[0172] C=Inorm′·S′

[0173] X=C·(1-|(H′×6)mod2-1|)

[0174]

[0175] Among them, (R, G, B) represents the RGB value of the target contour image in RGB form after conversion, R', G', B', C, X and m are intermediate calculation variables, Inorm' represents the enhanced I array, S' represents the enhanced S array, and H' represents the corrected H array.

[0176] In one embodiment, the equalization processing module 502 is specifically used to divide the image corresponding to the normalized I array into a plurality of I channel grid blocks according to a preset size; calculate the minimum pixel value, maximum pixel value, grid block mean value and grid block standard deviation value of each I channel grid block; adjust the normalized I array according to the minimum pixel value, maximum pixel value, grid block mean value and grid block standard deviation value, and set the adjusted I array as the enhanced I array, and the adjustment method includes:

[0177]

[0178] Among them, Inorm′ n Indicates the adjusted nth pixel value, Inormn Represents the nth pixel value of the image corresponding to the normalized I array, I min Indicates the minimum pixel value, I max represents the maximum pixel value, and L represents the maximum range of pixel values;

[0179] Calculate the H channel local mean and H channel local standard deviation of the preset area of ​​the normalized H array, as well as the H channel global mean and H channel global standard deviation of the normalized H array; correct the normalized H array according to the H channel local mean, the H channel local standard deviation, the H channel global mean and the H channel global standard deviation, and set the corrected H array as the corrected H array, and the correction method includes:

[0180]

[0181] Among them, H′ n Indicates the nth value of the H array, H norm，n Represents the normalized nth value of the H array, μ H represents the local mean of the H channel, σ H Represents the local standard deviation of the H channel, Represents the global standard deviation of the H channel, represents the global mean of H channel;

[0182] Calculate the local mean value of the S channel and the local standard deviation value of the S channel in the preset area of ​​the normalized S array; adjust the normalized S array according to the local mean value of the S channel and the local standard deviation value of the S channel, and set the adjusted S array as the enhanced S array, and the adjustment method includes:

[0183]

[0184] Among them, α and β represent adaptive adjustment parameters, S′ n Indicates enhancing the nth value of the S array, S norm,n Represents the normalized nth value of the S array, μ S represents the local mean of the S channel, σ S Indicates the local standard deviation of the S channel.

[0185] In one embodiment, the image output module 504 is also used to obtain a set of images with clear material stacking contours and feeding text information corresponding to the set of images with clear material stacking contours, wherein the set of images with clear material stacking contours is a collection of images with clear material stacking contours at the material inlet; a histogram equalization algorithm is used to preprocess each image in the set of images with clear material stacking contours, and a set of equalized contour images is generated; a preset number of Gaussian noises are added to each image in the set of equalized contour images, and a set of fuzzy contour images is generated; a contour enhancement neural network model is used to fuse each image in the set of fuzzy contour images with the corresponding feeding text information, and a fused image set is generated; a Bayesian denoising algorithm is used to denoise each image in the fused image set, and a denoised contour image set is generated; the image data difference between each image in the denoised contour image set and the corresponding image in the set of equalized contour images is calculated; if the image data difference is within a preset range, the training of the contour recognition neural network model is completed.

[0186] In one embodiment, the image output module 504 is also used to encode the feed text information and generate feed text encoding data; fuse the data of each image in the fuzzy contour image set with the corresponding feed text encoding data to generate image-text fusion data; use a contour enhancement neural network model to convert the image-text fusion data into a fused image, and generate a fused image set.

[0187] In one embodiment, the image output module 504 is also used to obtain the feeding text information corresponding to the original contour image, fuse the clear contour image with the feeding text information corresponding to the original contour image, and generate edge prediction raw data; input the edge prediction raw data into the trained edge prediction neural network model, and output the predicted edge contour image.

[0188] In one embodiment, the image output module 504 is also used to construct an edge prediction neural network model, which includes six convolutional layers and five pooling layers; input the edge prediction raw data into the edge prediction neural network model for feature extraction, and output the predicted edge image; construct a loss function according to the material type, and optimize the edge prediction neural network model according to the loss function to generate a trained edge prediction neural network model.

[0189] In one embodiment, the image output module 504 is also used to upsample the data output by the seventh convolution layer by 4 times to generate the upsampled data of the seventh convolution layer, and upsample the data output by the fourth pooling layer by 2 times to generate the upsampled data of the fourth pooling layer; fuse the upsampled data of the seventh convolution layer, the upsampled data of the fourth pooling layer and the data output by the third pooling layer to generate the convolutional pooling fusion data; upsample the convolutional pooling fusion data by 8 times to output the predicted edge image.

[0190] The powder pneumatic conveying image processing device provided in the embodiment of the present application can be applied to the powder pneumatic conveying image processing method provided in the above embodiment. For relevant details, refer to the above method embodiment. The implementation principle and technical effect are similar and will not be repeated here.

[0191] It should be noted that: the powder pneumatic conveying image processing device provided in the embodiment of the present application only uses the division of the above-mentioned functional modules / functional units as an example when performing powder pneumatic conveying image processing. In actual applications, the above-mentioned functions can be assigned to different functional modules / functional units as needed, that is, the internal structure of the powder pneumatic conveying image processing device is divided into different functional modules / functional units to complete all or part of the functions described above. In addition, the implementation method of the powder pneumatic conveying image processing method provided in the above method embodiment and the implementation method of the powder pneumatic conveying image processing device provided in this embodiment belong to the same concept. The specific implementation process of the powder pneumatic conveying image processing device provided in this embodiment is detailed in the above method embodiment, which will not be repeated here.

[0192] The embodiment of the present application also discloses a computer device.

[0193] Specifically, if Figure 6 As shown, the computer device can be a computer device such as a desktop computer, a laptop computer, a PDA, and a cloud server. The computer device may include, but is not limited to, a processor and a memory. The processor and the memory may be connected via a bus or otherwise. The processor may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, graphics processors (GPU), embedded neural network processors (NPU) or other dedicated deep learning coprocessors, discrete gates or transistor logic devices, discrete hardware components and other chips, or a combination of the above-mentioned various chips.

[0194] As a non-transient computer-readable storage medium, the memory can be used to store non-transient software programs, non-transient computer executable programs and modules, such as program instructions / modules corresponding to the methods in the above-mentioned embodiments of the present application. The processor executes various functional applications and data processing of the processor by running the non-transient software programs, instructions and modules stored in the memory, that is, the method in the above-mentioned method implementation is realized. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0195] The embodiment of the present application also discloses a computer-readable storage medium.

[0196] Specifically, a computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the method in the above-mentioned method implementation is implemented. Those skilled in the art will understand that the implementation of all or part of the process in the above-mentioned implementation method of the present application can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and the program, when executed, may include the process of the implementation of the above-mentioned methods. Among them, the storage medium may be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory (Flash Memory), a hard disk (Hard Disk Drive, abbreviated as: HDD) or a solid-state drive (SSD), etc.; the storage medium may also include a combination of the above-mentioned types of memory.

[0197] This specific embodiment is merely an explanation of the present invention and is not a limitation of the present invention. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed. However, as long as they are within the scope of the claims of the present invention, they are protected by the patent law.

Claims

1. A method for image processing of powder pneumatic conveying, characterized in that: The method comprises: Acquire an image of the material accumulation contour of the feed inlet and corresponding feeding text information, and set the image of the material accumulation contour of the feed inlet as an original contour image; Adopting an adaptive contrast enhancement fusion color correction algorithm to preprocess the original contour image and generate a target contour image; Adding a preset number of Gaussian noises to the target contour image to generate a noise contour image; The noise contour image and the corresponding feeding text information are input into the trained contour enhancement neural network model, and a clear contour image is output after Bayesian denoising.

2. The method according to claim 1, characterized in that: The original contour image is a color image in RGB format, and the method of preprocessing the original contour image using an adaptive contrast enhancement fusion color correction algorithm to generate a target contour image includes: The original contour image in RGB format is converted into the original contour image in HSI format, wherein the original contour image in HSI format includes an H channel, an S channel, and an I channel. The conversion method includes: H=360°-H(if B>G) Among them, R, G, B represent the data of R channel, G channel and B channel of the original contour image in RGB format, respectively, and H, S, I represent the data of H channel, S channel and I channel of the original contour image in HSI format after conversion, respectively; The data of the H channel, the S channel and the I channel are respectively normalized to generate a normalized H array, a normalized S array and a normalized I array. The calculation method of the data normalization of the H channel, the S channel and the I channel includes: Among them, H norm represents the normalized H array, S norm represents the normalized S array, I norm represents the normalized I array, I max Indicates the maximum value in the I channel data; Adopting an adaptive contrast enhancement algorithm, a hue correction algorithm, and a saturation enhancement algorithm to process the normalized I array, the normalized H array, and the normalized S array, respectively, and generating an enhanced I array, a corrected H array, and an enhanced S array, respectively; The corrected H array, the enhanced S array and the enhanced I array are merged to generate a corrected HSI image; The corrected HSI image is converted into an image in RGB format, and the target contour image is generated. The conversion method includes: (R, G, B) = ((R′+m)×255, (G′+m)×255, (B′+m)×255), in, C=Inorm′.S′ X=C·(1-|(H′×6)mod 2-1|) Wherein, (R, G, B) represents the RGB value of the target contour image in the converted RGB form, R', G', B', C, X and m are intermediate calculation variables, Inorm' represents the enhanced I array, S' represents the enhanced S array, and H' represents the corrected H array.

3. The method according to claim 2, characterized in that: The adopting of the adaptive contrast enhancement algorithm, the hue correction algorithm and the saturation enhancement algorithm to process the normalized I array, the normalized H array and the normalized S array respectively, and generating an enhanced I array, a corrected H array and an enhanced S array respectively comprises: Dividing the image corresponding to the normalized I array into a plurality of I channel grid blocks according to a preset size; Calculate the minimum pixel value, maximum pixel value, grid block mean value and grid block standard deviation value of each of the I channel grid blocks; The normalized I array is adjusted according to the minimum pixel value, the maximum pixel value, the grid block mean value, and the grid block standard deviation value, and the adjusted I array is set as the enhanced I array, and the adjustment method includes: Among them, Inorm′ n Indicates the adjusted nth pixel value, Inorm n Represents the nth pixel value of the image corresponding to the normalized I array, I min represents the minimum pixel value, I max represents the maximum pixel value, and L represents the maximum range of pixel values; Calculate the H channel local mean and H channel local standard deviation of the preset area of ​​the normalized H array, and the H channel global mean and H channel global standard deviation of the global normalized H array; The normalized H array is corrected according to the H channel local mean, the H channel local standard deviation, the H channel global mean, and the H channel global standard deviation, and the corrected H array is set as the corrected H array, and the correction method includes: Among them, H′ n Indicates the nth value of the calibration H array, H norm，n represents the nth value of the normalized H array, μ H represents the local mean of the H channel, σ H Represents the local standard deviation of the H channel, σH global Indicates the global standard deviation of the H channel, μH global represents the global mean of H channel; Calculating the S channel local mean and the S channel local standard deviation of a preset area of ​​the normalized S array; The normalized S array is adjusted according to the local mean value of the S channel and the local standard deviation value of the S channel, and the adjusted S array is set as the enhanced S array. The adjustment method includes: Among them, α and β represent adaptive adjustment parameters, S′ n represents the nth value of the enhanced S array, S norm，n represents the nth value of the normalized S array, μ S represents the local mean of the S channel, σ S Indicates the local standard deviation of the S channel.

4. The method according to claim 1, characterized in that: The training method of the contour enhancement neural network model includes: Acquire a material accumulation clear outline image set and feeding text information corresponding to the material accumulation clear outline image set, wherein the material accumulation clear outline image set is a collection of material accumulation clear outline images at a material inlet; Using a histogram equalization algorithm to pre-process each image in the material accumulation clear contour image set, and generate a balanced contour image set; Adding a preset number of Gaussian noises to each image in the equalized contour image set, and generating a fuzzy contour image set; Using a contour enhancement neural network model to fuse each image in the fuzzy contour image set with the corresponding feed text information, and generate a fused image set; De-noising each image in the fused image set using a Bayesian denoising algorithm to generate a de-noised contour image set; Calculate the image data difference between each image in the denoised contour image set and the corresponding image in the equalized contour image set; If the image data difference is within a preset range, the contour recognition neural network model training is completed.

5. The method according to claim 4, characterized in that: The step of using a contour enhancement neural network model to fuse each image in the fuzzy contour image set with the corresponding feed text information and generating a fused image set comprises: Encoding the feeding text information and generating feeding text encoding data; Merging the data of each image in the fuzzy contour image set with the corresponding input text encoding data to generate image-text fusion data; The image-text fusion data is converted into a fused image by using a contour enhancement neural network model, and the fused image set is generated.

6. The method according to claim 1, characterized in that: After outputting the clear contour image, the method further comprises: Acquire the feeding text information corresponding to the original contour image, fuse the clear contour image with the feeding text information corresponding to the original contour image, and generate edge prediction original data; The edge prediction raw data is input into the trained edge prediction neural network model, and the predicted edge contour image is output.

7. The method according to claim 6, characterized in that: The training method of the edge prediction neural network model includes: Constructing an edge prediction neural network model, wherein the edge prediction neural network model includes six convolutional layers and five pooling layers; Inputting the edge prediction raw data into the edge prediction neural network model for feature extraction, and outputting a predicted edge image; A loss function is constructed according to the material type, and the edge prediction neural network model is optimized according to the loss function to generate a trained edge prediction neural network model.

8. The method according to claim 7, characterized in that: The step of inputting the edge prediction raw data into the edge prediction neural network model for feature extraction and outputting the predicted edge image comprises: The data output by the seventh convolutional layer is upsampled by 4 times to generate the upsampled data of the seventh convolutional layer, and the data output by the fourth pooling layer is upsampled by 2 times to generate the upsampled data of the fourth pooling layer; Fusing the upsampled data of the seventh convolutional layer, the upsampled data of the fourth pooling layer, and the data output by the third pooling layer to generate convolutional pooling fusion data; The convolution pooling fusion data is upsampled 8 times, and the predicted edge image is output.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executes the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 8.

Images