Rotary kiln working condition identification method, device, equipment and medium

Abstract

The present invention provides a method, device, equipment and medium for identifying the working condition of a rotary kiln. The method comprises: obtaining a combustion image taken from a fire detection port of a rotary kiln toward the inside of a rotary kiln during the combustion process of the rotary kiln, and inputting the combustion image into a preset image classification model; extracting a feature image of the combustion image, and performing circular position encoding on the feature image with a position corresponding to the axis of the rotary kiln in the feature image as the center of the circle, so as to obtain circular position encoding information at each position in the feature image; dividing the feature image into a plurality of image blocks, and adding each circular position encoding information to the image block at the corresponding position, so as to adjust the weight of each image block; performing self-attention processing on the plurality of image blocks based on the adjusted weights, so as to obtain corresponding self-attention features, and performing image classification based on the self-attention features, so as to obtain the working condition identification result during the combustion process of the rotary kiln, so as to improve the accuracy of the working condition identification of the rotary kiln.

Description

Technical Field

[0001] The present disclosure relates to the field of metallurgical equipment modeling and application technology, and in particular to a method, device, equipment and medium for identifying the working condition of a rotary kiln. Background Art

[0002] The rotary kiln is a large-scale sintering equipment in the industrial field and a large-scale rotary thermal equipment widely used in industrial production. The working conditions of the rotary kiln are generally divided into under-burning, normal burning and over-burning according to the production efficiency. Normal burning is the ideal working condition, while under-burning or over-burning needs to be adjusted in time to restore to the normal burning state. Therefore, accurately identifying the working conditions of the rotary kiln is important for improving the combustion efficiency of the rotary kiln.

[0003] At present, manually observing the combustion conditions of the rotary kiln to judge the working condition is easily affected by the internal structure of the rotary kiln and the difference in workers' experience. In the related technology, the overall image information of the rotary kiln is gradually obtained for working condition recognition, but due to the structural characteristics of the rotary kiln, the problem of incomplete feature extraction is prone to occur, thereby reducing the accuracy of working condition recognition. Summary of the invention

[0004] The main purpose of the embodiments of the present disclosure is to provide a method, device, equipment and medium for identifying the working condition of a rotary kiln, which can improve the accuracy of identifying the working condition of the rotary kiln.

[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present disclosure provides a method for identifying the working condition of a rotary kiln, comprising:

[0006] Acquire a combustion image captured from a fire detection port of the rotary kiln toward the interior of the rotary kiln during the combustion process of the rotary kiln, and input the combustion image into a preset image classification model;

[0007] Extracting a characteristic image of the combustion image, and taking a position in the characteristic image corresponding to the axis of the rotary kiln as the center of a circle, performing circular position coding on the characteristic image to obtain circular position coding information at each position in the characteristic image;

[0008] Dividing the feature image into a plurality of image blocks, and adding each circular position encoding information to the image block at a corresponding position to adjust the weight of each image block;

[0009] Based on the adjusted weights, self-attention processing is performed on the plurality of image blocks to obtain corresponding self-attention features, and image classification is performed based on the self-attention features to obtain working condition recognition results during the combustion process of the rotary kiln.

[0010] In some embodiments, dividing the feature image into a plurality of image blocks, and adding each circular position encoding information to the image block at a corresponding position to adjust the weight of each image block includes:

[0011] Performing convolution processing on the feature image to adjust the number of channels to obtain a first convolution image;

[0012] The first convolution image is expanded and divided into a plurality of image blocks, and each circular position encoding information is added to the image block at a corresponding position to adjust the weight of each image block.

[0013] In some embodiments, the image classification based on the self-attention feature is performed to obtain the working condition recognition result during the combustion process of the rotary kiln, including:

[0014] Folding each of the self-attention features to obtain a self-attention image, and performing convolution processing on the self-attention image to restore the number of channels to obtain a second convolution image;

[0015] Splicing the second convolution image with the feature image to obtain a spliced ​​image, and performing convolution processing on the spliced ​​image to restore the number of channels to obtain a third convolution image;

[0016] Image classification is performed based on the third convolution image to obtain an operating condition recognition result during the combustion process of the rotary kiln.

[0017] In some embodiments, the image classification model includes a plurality of mobile visual transformation blocks at different stages, the feature image is used as input data of any one of the mobile visual transformation blocks, and the third convolution image is output data of the corresponding mobile visual transformation block;

[0018] The image classification is performed based on the third convolution image to obtain the working condition recognition result of the rotary kiln combustion process, including:

[0019] The third convolution image is used as a feature image of the next stage, and is input into the moving visual transformation block of the next stage, until the moving visual transformation block of the last stage outputs the corresponding third convolution image, wherein the moving visual transformation blocks at multiple different stages share the same circular position code;

[0020] Global pooling and full connection processing are performed based on the last third convolution image to obtain a fully connected feature, and image classification is performed based on the fully connected feature to obtain an operating condition recognition result during the combustion process of the rotary kiln.

[0021] In some embodiments, the image classification model is trained by the following steps, including:

[0022] Acquire a plurality of sample images taken from a fire detection port of the rotary kiln toward the interior of the rotary kiln during the combustion process of the rotary kiln;

[0023] Marking the working condition of each of the sample images to obtain a working condition label;

[0024] Inputting each of the sample images into an initial image classification model to obtain each sample working condition recognition result;

[0025] The initial image classification model is trained based on the sample working condition recognition results and the corresponding working condition labels until a training stop condition is met, thereby obtaining the trained image classification model.

[0026] In some embodiments, the acquiring of a plurality of sample images taken from a fire detection port of the rotary kiln toward the interior of the rotary kiln during the combustion process of the rotary kiln includes:

[0027] Acquire a sample video captured from a fire detection port of the rotary kiln toward the interior of the rotary kiln during the combustion process of the rotary kiln;

[0028] The sample video is sliced ​​to obtain a plurality of video segments, and a video frame with a preset duration is selected from each of the video segments as a sample image.

[0029] In some embodiments, labeling the working condition of each of the sample images to obtain a working condition label includes:

[0030] Acquire the working data of the rotary kiln corresponding to each of the sample images, wherein the working data includes at least one of the temperature at the head of the rotary kiln, the temperature at the tail of the rotary kiln, the frequency of the disc feeder, the frequency of the main motor, and the outlet pressure of the fan;

[0031] The corresponding working condition of the rotary kiln is determined based on each of the working data, and the corresponding sample images are marked based on each of the working conditions to obtain a working condition label.

[0032] In some embodiments, taking the position in the characteristic image corresponding to the axis of the rotary kiln as the center of the circle, performing circular position coding on the characteristic image to obtain circular position coding information at each position in the characteristic image includes:

[0033] Taking the position in the characteristic image corresponding to the axis of the rotary kiln as the center of the circle, establishing Cartesian coordinate information of each position in the characteristic image;

[0034] The Cartesian coordinate information at each position in the feature image is converted into polar coordinate information, and circular position coding information at each position in the feature image is obtained based on the polar coordinate information.

[0035] In some embodiments, obtaining circular position coding information at each position in the feature image based on the polar coordinate information includes:

[0036] Performing non-uniform integer quantization processing based on the size of the Cartesian coordinate information at each position in the feature image to obtain processed polar coordinate information;

[0037] The circular position coding information at each position in the feature image is obtained based on the processed polar coordinate information.

[0038] To achieve the above-mentioned purpose, a second aspect of the embodiment of the present disclosure provides a working condition identification device for a rotary kiln, comprising:

[0039] An image acquisition module is used to acquire a combustion image taken from a fire detection port of the rotary kiln toward the inside of the rotary kiln during the combustion process of the rotary kiln, and input the combustion image into a preset image classification model;

[0040] A circular position coding module is used to extract the characteristic image of the combustion image, and to perform circular position coding on the characteristic image with the position corresponding to the axis of the rotary kiln in the characteristic image as the center of the circle, so as to obtain circular position coding information at each position in the characteristic image;

[0041] A coding embedding module, used for dividing the feature image into a plurality of image blocks, and adding the circular position coding information to the image block at the corresponding position to adjust the weight of each image block;

[0042] The working condition recognition module is used to perform self-attention processing on the multiple image blocks based on the adjusted weights to obtain corresponding self-attention features, and perform image classification based on the self-attention features to obtain working condition recognition results during the combustion process of the rotary kiln.

[0043] To achieve the above-mentioned purpose, the third aspect of the embodiment of the present disclosure proposes an electronic device, which includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements the rotary kiln operating condition identification method described in the embodiment of the first aspect.

[0044] To achieve the above-mentioned purpose, the fourth aspect of the embodiment of the present disclosure proposes a storage medium, which is a computer-readable storage medium, and the storage medium stores a computer program. When the computer program is executed by a processor, the method for identifying the working condition of the rotary kiln described in the embodiment of the first aspect above is implemented.

[0045] The working condition identification method, device, equipment and medium of the rotary kiln proposed in the embodiments of the present disclosure can be applied to the working condition identification device of the rotary kiln. By executing the working condition identification method of the rotary kiln, the combustion image obtained by shooting from the fire detection port of the rotary kiln to the inside of the rotary kiln during the combustion process of the rotary kiln can be first obtained, and the combustion image can be input into the preset image classification model. Since the rotary kiln has the physical property of a rotating circle, the combustion image obtained from the fire detection port is a partial image of the circular cross-section of the rotary kiln. Therefore, after extracting the feature image of the combustion image, the feature image can be circularly encoded with the position corresponding to the axis of the rotary kiln in the feature image as the center of the circle to obtain the circular position encoding information at each position in the feature image. Then, the feature image is divided into multiple image blocks, and each circular position encoding information is added to the image block at the corresponding position. The weight of each image block can be adjusted, and the position information expression capability of the material inside the rotary kiln body can be enhanced to better extract features. Finally, by performing self-attention processing on multiple image blocks based on the adjusted weights to obtain the corresponding self-attention features, and performing image classification based on the self-attention features, the operating condition recognition results of the rotary kiln combustion process can be obtained, thereby improving the accuracy of rotary kiln operating condition recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a schematic flow chart of a method for identifying the working condition of a rotary kiln provided in an embodiment of the present disclosure;

[0047] Figure 2 yes Figure 1 Step S103 further includes a flow chart;

[0048] Figure 3 is a structural diagram of an image classification model provided by an embodiment of the present disclosure;

[0049] Figure 4 yes Figure 1 Step S104 further includes a flow chart;

[0050] Figure 5 yes Figure 4 Step S303 further includes a flow chart;

[0051] Figure 6 is a flowchart of the image classification model training process provided by the embodiment of the present disclosure;

[0052] Figure 7 yes Figure 6 Step S501 further includes a flow chart;

[0053] Figure 8 yes Figure 6 Step S502 further includes a flow chart;

[0054] Fig. 9 is a schematic diagram of a process of obtaining a sample image and a corresponding working condition label provided by an embodiment of the present disclosure;

[0055] Fig.10 is a schematic diagram of a process of obtaining a sample image and a corresponding working condition label provided by an embodiment of the present disclosure;

[0056] Fig.11 is a flowchart of a circular position encoding process provided by an embodiment of the present disclosure;

[0057] Fig.12 yes Fig.10 Step S802 further includes a flow chart;

[0058] Fig.13 It is a functional module schematic diagram of a rotary kiln operating condition identification device provided by an embodiment of the present disclosure;

[0059] Fig.14 It is a schematic diagram of the hardware structure of the electronic device provided by the embodiment of the present disclosure. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present disclosure more clear, the present disclosure is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present disclosure and are not used to limit the present disclosure.

[0061] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present disclosure belongs. The terms used herein are only for the purpose of describing the embodiments of the present disclosure and are not intended to limit the present disclosure.

[0062] First, some nouns involved in this disclosure are analyzed:

[0063] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robots, language recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0064] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0065] Vision Transformer (ViT) is a popular deep network structure in the field of computer vision. It is the first time that the Transformer model has been applied to the field of computer vision. Its basic component is the self-attention module.

[0066] Convolutional Neural Networks (CNN) are a type of feedforward neural networks that include convolution calculations and have a deep structure. They are one of the representative algorithms of deep learning.

[0067] Mobile Vision Transformer (MobileViT) is a lightweight vision model based on the ViT architecture, designed to be suitable for mobile devices and embedded systems. MobileViT combines the advantages of CNN and ViT. It uses the lightweight and efficient characteristics of CNN and the self-attention mechanism and global vision of Transformer to build a lightweight, general and mobile device-friendly network model.

[0068] The Mobile ViT Block (MobileViT Block), also known as the MobileViT block, is the core component of MobileViT. It uses convolutional layers to encode local spatial information and projects tensors into high-dimensional space through point convolutions. It aims to model local and global information in the input tensor with fewer parameters.

[0069] A Distributed Control System (DCS) is an automated control system consisting of distributed controllers and field devices. It can monitor and control all aspects of industrial production in real time. It consists of multiple control units scattered in different areas or equipment, which are connected through a network and work together to complete the automated control of industrial processes.

[0070] The rotary kiln is a large-scale sintering equipment in the industrial field and a large-scale rotary thermal equipment widely used in industrial production. For example, the rotary kiln includes the zinc oxide rotary kiln, which is a large-scale sintering equipment in the industrial field. It is mainly used to process zinc leaching residues in hydrometallurgy. By feeding a mixture of zinc leaching residue and coke into the kiln body and controlling the temperature and operating conditions, the mixture undergoes an oxidation-reduction reaction in the rotating high-temperature kiln body, and finally obtains zinc oxide. The working conditions of the rotary kiln are generally divided into under-burning, normal burning, and over-burning according to production efficiency. Normal burning is the ideal working condition, while under-burning or over-burning needs to be adjusted in time to restore to the normal burning state. Therefore, accurately identifying the working conditions of the rotary kiln plays an important role in improving the combustion efficiency of the rotary kiln.

[0071] At present, manually observing the combustion conditions of the rotary kiln to judge the working condition is easily affected by the internal structure of the rotary kiln and the difference in workers' experience. Some key local information is not easy to observe, which often leads to errors in the judgment of the working condition. In the related technology, the overall image information of the rotary kiln is gradually obtained for working condition recognition. However, due to the structural characteristics of the rotary kiln, it is easy to have incomplete feature extraction, thereby reducing the accuracy of working condition recognition.

[0072] Based on this, the embodiments of the present disclosure provide a method, device, equipment and medium for identifying the working condition of a rotary kiln, which can improve the accuracy of identifying the working condition of the rotary kiln.

[0073] The working condition identification method of the rotary kiln in the embodiment of the present disclosure can be illustrated by the following embodiment.

[0074] The embodiments of the present disclosure can acquire and process relevant data based on artificial intelligence technology.

[0075] The working condition identification method of the rotary kiln provided in the embodiment of the present disclosure can be applied to the terminal, can also be applied to the server side, and can also be software running in the terminal or the server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, and can also be configured as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the working condition identification method of the rotary kiln, etc., but is not limited to the above forms.

[0076] The present disclosure can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present disclosure can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present disclosure can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0077] It should be noted that in various specific embodiments of the present disclosure, when obtaining the combustion image or the working data of the rotary kiln for related processing, the user's permission or consent will be obtained first. Moreover, the collection, use and processing of these data will comply with relevant laws, regulations and standards. In addition, when the embodiment of the present disclosure needs to obtain the combustion image or the working data of the rotary kiln, the user's separate permission or consent can be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary combustion image or rotary kiln working data for the normal operation of the embodiment of the present disclosure can be obtained.

[0078] Figure 1 is a flow chart of a method for identifying the working condition of a rotary kiln provided in an embodiment of the present disclosure, Figure 1 The method may include but is not limited to steps S101 to S104.

[0079] Step S101, obtaining a combustion image taken from a fire detection port of the rotary kiln toward the inside of the rotary kiln during the combustion process of the rotary kiln, and inputting the combustion image into a preset image classification model;

[0080] Step S102, extracting a characteristic image of the combustion image, and performing circular position coding on the characteristic image with the position corresponding to the axis of the rotary kiln in the characteristic image as the center of the circle, to obtain circular position coding information at each position in the characteristic image;

[0081] Step S103, dividing the feature image into a plurality of image blocks, and adding each circular position encoding information to the image block at the corresponding position to adjust the weight of each image block;

[0082] Step S104, performing self-attention processing on multiple image blocks based on the adjusted weights to obtain corresponding self-attention features, and performing image classification based on the self-attention features to obtain working condition recognition results during the rotary kiln combustion process.

[0083] With respect to the above step S101, the combustion image is an image captured from the fire detection port of the rotary kiln into the kiln by a specific shooting device during the combustion process of the rotary kiln. The fire detection port is a specially designed opening on the rotary kiln for observing the combustion conditions in the kiln. Exemplarily, the combustion image in the embodiment of the present disclosure may be captured by a high-resolution camera or other imaging device located on the fire detection port, and the shooting process may be real-time or periodic shooting to ensure that the acquired combustion image can accurately reflect the current combustion state.

[0084] Exemplarily, the rotary kiln in the embodiment of the present disclosure is a zinc oxide rotary kiln. On the premise of meeting the requirements of the embodiment of the present disclosure, the rotary kiln can also be of other types, for example, it can be a metal magnesium rotary kiln, a bauxite rotary kiln, etc., and the embodiment of the present disclosure does not make specific limitations on this.

[0085] Furthermore, the combustion image can contain a variety of information. For example, the combustion image can contain information such as the shape and color of the flame. The shape of the flame can reflect the stability and uniformity of the combustion in the kiln, while the color of the flame can provide information on the combustion temperature and oxygen content. The combustion image can also contain the state of the material in the kiln, such as the color, shape, and particle size of the material. These characteristics can reflect the degree of combustion and sintering effect of the material in the kiln. The combustion image can also include the overall environment in the kiln, such as the temperature of the kiln wall and the airflow conditions in the kiln. This information is helpful for analyzing the thermal process in the kiln.

[0086] The image classification model is a pre-trained classification model that can classify images. Specifically, the image classification model can receive combustion images taken from the fire detection port of the rotary kiln, and classify the working conditions based on these images to determine the combustion state in the rotary kiln, for example, to determine any one of the working conditions of under-burning, normal burning or over-burning. In the embodiment of the present disclosure, the image classification model is a ViT model, and further, it can also be a MobileViT model or other types of ViT models.

[0087] For the above step S102, after the combustion image is input into the image classification model, the image classification model is provided with a plurality of network layers, through which the input combustion image can be feature extracted to obtain a feature image, which contains key information in the combustion image but removes redundant or unnecessary details.

[0088] It should be noted that since the combustion image is obtained by the camera through the fire detection port of the rotary kiln, due to the size limitation of the fire detection port, the obtained image is a partial image of the circular cross-section of the rotary kiln, and since the placement position of the shooting camera is not on the axis line of the kiln body, the embodiment of the present disclosure provides a method for determining the center of the circular position code according to the acquired combustion image of the rotary kiln, and determining the distance and angle quantization method in the circular position code design according to the characteristics of the rotary kiln.

[0089] Specifically, the disclosed embodiment can find the position corresponding to the axis of the rotary kiln on the feature image. This point will be used as the center of the circular position code. The position of the center of the circle can be determined by identifying the image, or it can be pre-configured. The disclosed embodiment does not impose specific restrictions on this. After the center of the circle is determined, each position on the feature image can be circularly coded. The purpose of circular position coding is to embed the position information, that is, the position of each pixel relative to the center of the circle, into the feature image. Specifically, the disclosed embodiment can calculate the distance and angle from each pixel to the center of the circle, and then convert these distance and angle information into coded values. The coded values ​​can be continuous, such as using polar coordinates, or discrete, such as dividing the circle into multiple sectors, each sector is assigned a unique code. After the above steps, each position in the feature image will have a corresponding circular position coding information.

[0090] Since the circular position coding in the embodiment of the present disclosure is a position coding method designed in consideration of the circular physical properties of the rotary kiln, this method can enhance the ability to express the position information of the material inside the rotary kiln body, which is crucial to improving the accuracy of operating condition identification. In particular, when there is uneven combustion or material distribution inside the rotary kiln, the position information can help the model to more accurately judge the operating condition.

[0091] For the above step S103, in the embodiment of the present disclosure, it is necessary to divide the feature image into multiple smaller image blocks, namely, image patches. These image blocks can be rectangular areas of fixed size or other shapes, and the number and size of the divided image blocks can be adjusted according to the specific application scenarios and model requirements.

[0092] After dividing the image blocks, the circular position encoding information needs to be added to the image blocks at the corresponding positions. Specifically, for each image block, its position in the feature image needs to be found, and the corresponding circular position encoding information needs to be found based on this position. These position encoding information will be added to the image block as additional features for subsequent self-attention processing.

[0093] In the disclosed embodiment, by adding circular position coding information to the image block, the weight of each image block is actually adjusted, and the weight reflects the importance of the image block in the material position information inside the rotary kiln. For example, the image block located in the center area of ​​the rotary kiln may contain more critical information, so it should be given a higher weight. Finally, each image block contains the original feature information and the added circular position coding information.

[0094] Exemplarily, in the disclosed embodiment, the feature image can be divided into image blocks of fixed size, each image block is encoded into a token, and the token carries sequence information, and then the circular position encoding information of the corresponding position is added to the corresponding token, so that the weight of each image block can be adjusted.

[0095] With respect to the above step S104, the self-attention mechanism is a powerful feature extraction method in deep learning, which allows the model to focus on all elements in the input sequence when processing each element, that is, the image block, and dynamically adjust the weights according to the correlation between them. In the embodiment disclosed in the present invention, the self-attention mechanism is applied to the image blocks after weight adjustment, each image block is regarded as a "query", and all image blocks are regarded as "keys" and "values". By calculating the similarity between the query and the key (usually using dot products or other metrics), an attention score can be obtained, which reflects the degree of association between the query and each key. Then, these attention scores are used to weight the values ​​and sum them to obtain a new self-attention feature, which represents the fusion of information from all image blocks and is adjusted according to the correlation between them.

[0096] After the self-attention features are extracted, they are used for image classification. For example, the dimension of the self-attention features can be reduced to a fixed-size vector, which is then passed to a fully connected layer, which maps the classification embedding to the probability distribution of the operating condition category. This distribution is normalized by the softmax function to ensure that the sum of the probabilities of all categories is 1. Finally, the category with the highest probability is selected as the operating condition recognition result. Ultimately, the image classification model can output the operating condition recognition result of the rotary kiln combustion process, that is, one of underburning, normal burning, or overburning.

[0097] In summary, the embodiment of the present disclosure, through steps S101 to S104, can execute the working condition identification method of the rotary kiln, and can first obtain the combustion image taken from the fire detection port of the rotary kiln to the inside of the rotary kiln during the combustion process of the rotary kiln, and input the combustion image into a preset image classification model. Since the rotary kiln has the physical property of a rotating circle, the combustion image obtained from the fire detection port is a partial image of the circular cross-section of the rotary kiln. Therefore, after extracting the feature image of the combustion image, the feature image can be circularly position encoded with the position corresponding to the axis of the rotary kiln in the feature image as the center of the circle to obtain the circular position encoding information at each position in the feature image. Subsequently, the feature image is divided into multiple image blocks, and each circular position encoding information is added to the image block at the corresponding position. The weight of each image block can be adjusted, and the position information expression capability of the material inside the rotary kiln body can be enhanced to better extract features. Finally, by performing self-attention processing on multiple image blocks based on the adjusted weights to obtain the corresponding self-attention features, and performing image classification based on the self-attention features, the operating condition recognition results of the rotary kiln combustion process can be obtained, thereby improving the accuracy of rotary kiln operating condition recognition.

[0098] See also Figure 2 , Figure 2 yes Figure 1 In some embodiments, step S103 may include steps S201 to S202:

[0099] Step S201, performing convolution processing on the feature image to adjust the number of channels to obtain a first convolution image;

[0100] Step S202, expanding the first convolution image, dividing it into a plurality of image blocks, and adding each circular position encoding information to the image block at the corresponding position to adjust the weight of each image block.

[0101] In some embodiments, the image classification model is provided with a moving visual transformation block, and the input of the moving visual transformation block is a feature image. Figure 3 As shown, Figure 3A structural schematic diagram of an image classification model provided by an embodiment of the present disclosure is shown, wherein a mobile visual transformation block (ie, MobileViT block) is shown in the figure.

[0102] It should be noted that the feature image input by the mobile visual transformation block is a feature image of size H*W*C, and then the feature image needs to be convolved. A Local representations layer is set in the mobile visual transformation block, which consists of an n*n convolution layer and a 1*1 convolution layer. The 1*1 convolution layer is used to adjust the number of channels. Therefore, after the feature image is convolved by the Local representations layer, the first convolution image with a size of H*W*d can be obtained.

[0103] Then, following the Local representations layer is the Transformers as Convolution layer, which is an Unfold-Transformer-Fold structure. In this layer, the first convolution image can be unfolded to be divided into multiple image blocks, each image block is encoded as a token, and the token carries the order information, and then the circular position encoding information of the corresponding position is added to the corresponding token, so that the weight of each image block can be adjusted.

[0104] See also Figure 4 , Figure 4 yes Figure 1 In some embodiments, the image classification based on the self-attention feature in step S104 to obtain the working condition recognition result during the rotary kiln combustion process may include steps S301 to S303:

[0105] Step S301, folding each self-attention feature to obtain a self-attention image, and performing convolution processing on the self-attention image to restore the number of channels to obtain a second convolution image;

[0106] Step S302, splicing the second convolution image with the feature image to obtain a spliced ​​image, and performing convolution processing on the spliced ​​image to restore the number of channels to obtain a third convolution image;

[0107] Step S303: performing image classification based on the third convolution image to obtain a working condition recognition result during the combustion process of the rotary kiln.

[0108] In the above steps, if Figure 3As shown in the figure, the Transformers as Convolution layer is provided with an Unfold-Transformer-Fold structure. When the first convolution image is unfolded by Unfold and divided into multiple image blocks, and each circular position encoding information is added to the image block at the corresponding position to adjust the weight of each image block, the multiple image blocks can be self-attention processed by the Transformer layer. Finally, each self-attention feature is folded by Fold to obtain a self-attention image. Since the size of the self-attention image obtained after the Transformers as Convolution layer has not changed and is still H*W*d, it is necessary to perform convolution processing on the self-attention image to restore the number of channels, including adjusting the number of channels of the self-attention image to C through a 1*1 convolution layer, that is, the feature map size is transformed to H*W*C.

[0109] Afterwards, in the mobile visual transformation block, the generated second convolution image and the feature image will be spliced ​​through the shortcut branch to achieve feature fusion and enhance the feature expression ability. The spliced ​​image is then convolved to restore the number of channels, including processing through an n*n convolution layer to obtain the final output feature map, that is, the third convolution image. Finally, the third convolution image will be used as the output of the current mobile visual transformation block and processed through other subsequent network layers in the image classification model to perform image classification based on the third convolution image and obtain the working condition recognition result of the rotary kiln combustion process. Since the third convolution image has been fully extracted and integrated, it can more accurately reflect the combustion state and working condition of the rotary kiln.

[0110] It should be noted that, by introducing strategies such as the self-attention mechanism and feature splicing, the disclosed embodiment enables the image classification model to learn richer feature representations, thereby improving the generalization ability for different working conditions, so that the image classification model can make accurate working condition judgments when faced with new and unseen combustion images.

[0111] In some embodiments, Figure 3As shown, the image classification model is MobileViT, which includes multiple mobile visual transformation blocks of different stages. The feature image is used as the input data of any mobile visual transformation block, and the third convolution image is the output data of the corresponding mobile visual transformation block. MobileViT is set up with 5 stages. The first stage includes a MobileViNetV2 (MV2 for short), the second stage includes an MV2 with a downward arrow and an MV2, the third stage (No. 1 in the figure) includes an MV2 with a downward arrow and a MobileViT Block, the fourth stage (No. 2 in the figure) also includes an MV2 with a downward arrow and a MobileViT Block, and the fifth stage (No. 3 in the figure) also includes an MV2 with a downward arrow and a MobileViT Block.

[0112] Specifically, when the burning image is input into the image classification model, it is first convolved through a 3*3 convolution layer and downsampled by 2 times. The processed feature map is then input into an MV2, which may include a 1*1 convolution layer for dimensionality reduction, a 3*3 convolution layer, and a 1*1 convolution layer for dimensionality increase. Finally, the corresponding feature map is output to the next MV2. The MV2 structure marked with a downward arrow in the structure diagram represents the case where the stride is equal to 2, that is, downsampling is required. The layer structure includes a 1*1 convolution layer for dimensionality increase, a 3*3 convolution layer, and a 1*1 convolution layer for dimensionality reduction. Other structures in the subsequent image generation model will not be repeated.

[0113] See also Figure 5 , Figure 5 yes Figure 4 In some embodiments, step S303 may include steps S401 to S403:

[0114] Step S401, taking the third convolution image as the feature image of the next stage, and inputting it into the moving visual transformation block of the next stage, until the moving visual transformation block of the last stage outputs the corresponding third convolution image;

[0115] Among them, the mobile visual transformation blocks at different stages share the same circular position encoding;

[0116] Step S402, performing global pooling and full connection processing based on the last third convolution image to obtain a fully connected feature, and performing image classification based on the fully connected feature to obtain a working condition recognition result during the rotary kiln combustion process.

[0117] In the above steps, after processing the input feature image, the mobile visual transformation block in the third stage will output the third convolution image as the input data of the fourth stage. After the third convolution image is processed by the corresponding MV2, it will be processed as the feature image input by the mobile visual transformation block in the fourth stage. Similarly, in the fifth stage, after processing the third convolution image output from the fourth stage, the corresponding third convolution image will be output. Further, in the embodiment of the present disclosure, the third, fourth, and fifth stages (i.e. Figure 3 Therefore, when attention is repeatedly calculated within these three stages, the mobile visual transformation blocks within each stage share the same circular position encoding to reduce the amount of calculation.

[0118] After the mobile visual transformation block in the fifth stage outputs the corresponding third convolution image, in the embodiment of the present disclosure, the third convolution image can be first input into a 1*1 convolution layer for convolution processing. Then the corresponding feature map is input into the preset global pooling layer (Global pool) for global pooling processing, and the pooled features are input into the fully connected layer (Linear) for full connection processing to obtain full connection features, and image classification is performed based on the fully connected features to obtain the working condition recognition results during the combustion process of the rotary kiln.

[0119] It should be noted that the embodiment of the present disclosure only takes the setting of three mobile visual transformation blocks in the image classification module as an example for explanation. On the premise of meeting the requirements of the embodiment of the present disclosure, more stages of mobile visual transformation blocks can also be set to perform corresponding feature processing, and the embodiment of the present disclosure does not make any specific restrictions on this.

[0120] See also Figure 6 , Figure 6 is a flowchart of the image classification model training process provided by the embodiment of the present disclosure. In some embodiments, the image classification model is trained by the following steps, which may include steps S501 to S504:

[0121] Step S501, acquiring a plurality of sample images taken from a fire detection port of the rotary kiln toward the inside of the rotary kiln during the combustion process of the rotary kiln;

[0122] Step S502, marking the working condition of each sample image to obtain a working condition label;

[0123] Step S503, inputting each sample image into the initial image classification model to obtain each sample working condition recognition result;

[0124] Step S504, training the initial image classification model based on the sample working condition recognition results and the corresponding working condition labels until the training stop condition is met, thereby obtaining a trained image classification model.

[0125] In the above steps, the entire process for training the rotary kiln working condition recognition image classification model is described. During the training process, similar to the application process, the embodiment of the present disclosure first needs to obtain a large number of multiple sample images taken during the combustion process from the fire detection port of the rotary kiln as training samples. The multiple sample images cover various possible situations of the rotary kiln under different working conditions, including under-burning, normal burning and over-burning, etc. These images will serve as input data for model learning to help the model understand the relationship between the internal combustion state and working conditions of the rotary kiln.

[0126] Then, for each sample image, the embodiment of the present disclosure needs to mark its working condition. For example, the embodiment of the present disclosure can judge the working condition based on the features in the sample image, such as flame color, brightness, material form, etc., and assign a corresponding working condition label to each sample image, such as underburning, normal burning, and overburning. These working condition labels are supervisory information in the model training process, which are used to guide the model to learn the correct classification rules.

[0127] Next, the embodiment of the present disclosure inputs the acquired sample images and their corresponding working condition labels into the initial image classification model. The initial image classification model is an untrained or pre-trained neural network model. Specifically, the initial image classification model is MobileViT before training. The model processes each input image and outputs the sample working condition recognition results corresponding to each sample image.

[0128] During the training process, the disclosed embodiment can compare the difference between the sample working condition identification results output by the model and the working condition labels, and then adjust the parameters of the model according to this difference to reduce the error in future predictions, and perform multiple training iterations until a certain training stop condition is met, such as reaching a predetermined number of training rounds, the loss function value no longer significantly decreases, and the performance of the model on the validation set is stable. When these conditions are met, it can be considered that the model has fully learned how to identify working conditions from the combustion images of the rotary kiln, and can be used in actual working condition identification tasks. The image classification model obtained through training can more accurately identify the combustion images of the rotary kiln under different working conditions.

[0129] See also Figure 7 , Figure 7 yes Figure 6 In some embodiments, step S501 may include steps S601 to S602:

[0130] Step S601, obtaining a sample video shot from a fire detection port of the rotary kiln toward the inside of the rotary kiln during the combustion process of the rotary kiln;

[0131] Step S602: Slice the sample video to obtain multiple video segments, and select video frames with a preset duration in each video segment as sample images.

[0132] In the above steps, the process of obtaining the sample image from the sample video is described. Specifically, the embodiment of the present disclosure can obtain video data of the actual combustion process from the fire detection port of the rotary kiln to obtain the sample video, so as to collect enough and comprehensive data. At the same time, since the rotary kiln is continuously rotating, the video captured will be a dynamic process, including the combustion status of the material in the kiln at different positions and at different time points.

[0133] After obtaining the sample video, the embodiment of the present disclosure needs to cut the captured video into multiple shorter segments in chronological order. The length of these segments can be determined according to actual needs, and then select one or more representative images in each video segment as sample images. For example, an image at a specific time point in the segment (such as an image at the middle moment) can be selected, or the image with the best quality can be automatically selected based on a certain algorithm (such as an image quality assessment algorithm). For example, Fig. 9 As shown, Fig. 9 A schematic diagram of the process of obtaining sample images and corresponding working condition labels provided in an embodiment of the present disclosure is shown. In the embodiment of the present disclosure, the sample video can be sliced ​​according to a duration of every 5 minutes, and the middle image frame within the five minutes can be taken as the representative image of the five minutes, as the corresponding sample image.

[0134] Finally, the disclosed embodiment can obtain a large number of sample images from the combustion process of the rotary kiln. These images will serve as the basic data for subsequent image classification model training. The accuracy and comprehensiveness of these data will directly affect the training effect of the model and the final working condition recognition accuracy.

[0135] See also Figure 8 , Figure 8 yes Figure 6 In some embodiments, step S502 may include steps S701 to S702:

[0136] Step S701, obtaining the working data of the rotary kiln corresponding to each sample image;

[0137] The working data includes at least one of the temperature at the head of the rotary kiln, the temperature at the tail of the rotary kiln, the frequency of the disc feeder, the frequency of the main motor and the outlet pressure of the fan;

[0138] Step S702: determining the corresponding working condition of the rotary kiln based on each working data, and marking each corresponding sample image based on each working condition to obtain a working condition label.

[0139] In the above steps, a process of labeling sample images based on the working data of multiple rotary kilns to obtain corresponding working condition labels is described. Specifically, after obtaining the sample images, the working data of the rotary kiln corresponding to each sample image can be obtained in the embodiment of the present disclosure. For example, in the embodiment of the present disclosure, multiple DCS data can be obtained as working data through the DCS system, including at least one of the rotary kiln head temperature, rotary kiln tail temperature, disc feeder frequency, main motor frequency and fan outlet pressure.

[0140] Among them, the kiln head temperature of the rotary kiln is the temperature data of the kiln head part, which reflects the thermal state of the kiln head area; the kiln tail temperature of the rotary kiln is the temperature data of the kiln tail part, which helps to understand the thermal condition of the kiln tail area; the disc feeder frequency is the working frequency of the disc feeder, which reflects the speed or amount of material entering the rotary kiln; the main motor frequency is the operating frequency of the main motor that drives the rotary kiln to rotate, which is related to the speed of the rotary kiln; the fan outlet pressure is the outlet pressure of the fan used to control the gas flow in the kiln, which reflects the flow state of the gas in the kiln.

[0141] After obtaining the working data of each sample image, the working condition of the corresponding rotary kiln can be determined based on each working data. Fig. 9 As shown, the embodiment of the present disclosure can display corresponding sample images and corresponding work data to multiple excellent operators, obtain manually input discrimination data for each action data, and use the manually input discrimination data as the corresponding working condition label, thereby improving the accuracy of working condition labeling.

[0142] Furthermore, during the training process of the image classification model, the embodiment of the present disclosure can also pre-process the sample video and DCS data collected. Exemplarily, the pre-processing process includes data cleaning, the cleaning objects are measurement errors and other types of random errors; the pre-processing also includes outlier detection, which is to use statistical methods to identify and process outliers; the pre-processing also includes data balancing, where the video and DCS data that may be under-burned and over-burned working conditions will be over-sampled to increase the data under these two working conditions, so as to improve the imbalance of the original data volume under various working conditions that exists during model training.

[0143] See also Fig.10 , Fig.10 yes Figure 1In some embodiments, the step S102 further includes a flow chart of the step S102. In some embodiments, the step S102 takes the position corresponding to the axis of the rotary kiln in the feature image as the center of the circle, performs circular position coding on the feature image, and obtains the circular position coding information at each position in the feature image, which may include steps S801 to S802:

[0144] Step S801, taking the position in the characteristic image corresponding to the axis of the rotary kiln as the center of the circle, establishing Cartesian coordinate information of each position in the characteristic image;

[0145] Step S802 , converting the Cartesian coordinate information of each position in the feature image into polar coordinate information, and obtaining circular position coding information of each position in the feature image based on the polar coordinate information.

[0146] In the above steps, the detailed process of circular position coding is described. In the embodiment of the present disclosure, a zinc oxide rotary kiln is taken as an example. The zinc oxide rotary kiln is a cylinder. The combustion image is obtained by a camera through a fire detection port. Due to the size limitation of the fire detection port, the obtained image is a partial image of the circular cross-section of the rotary kiln. The image contains the material and flame information of the rotary kiln. Therefore, the circular position coding designed in the embodiment of the present disclosure is based on this physical feature of the rotary kiln.

[0147] Specifically, Fig.11 As shown, Fig.11 A flow chart of the circular position encoding process provided by an embodiment of the present disclosure is shown. In the circular position encoding process, it is necessary to determine the position of each pixel point in the feature image relative to the axis of the rotary kiln. Therefore, the embodiment of the present disclosure can use the position in the feature image corresponding to the axis of the rotary kiln as the center of the circle to establish Cartesian coordinate information at each position in the feature image, wherein the axis of the rotary kiln serves as the origin of the coordinate system, and each pixel point in the feature image has a corresponding (x, y) coordinate, so the position information of each pixel point can be simply obtained by reading the (x, y) value of each pixel point.

[0148] Next, the obtained Cartesian coordinate information can be converted to polar coordinate information. The polar coordinate system is a two-dimensional coordinate system in which any point can be located by the distance from the origin to the point (radius r) and the angle (θ) between the point and the positive x-axis. This conversion is particularly useful for the characteristic image of the rotary kiln because the kiln itself is circular and we are concerned with the position and angle of each point in the image relative to the axis of the kiln. By converting Cartesian coordinates to polar coordinates, the disclosed embodiments can more intuitively represent this position information and more easily encode this information based on the circular characteristics of the kiln.

[0149] Finally, after converting the Cartesian coordinate information at each position in the feature image into polar coordinate information, circular position coding information can be generated based on this information. Each pixel in the feature image can obtain a unique circular position coding information according to its position and angle relative to the kiln axis.

[0150] See also Fig.12 , Fig.12 yes Fig.10 In some embodiments, obtaining circular position coding information at each position in the feature image based on polar coordinate information in step S802 may include steps S901 to S902:

[0151] Step S901, performing non-uniform integer quantization processing based on the size of Cartesian coordinate information at each position in the feature image to obtain processed polar coordinate information;

[0152] Step S902, obtaining circular position coding information at each position in the feature image based on the processed polar coordinate information.

[0153] In the above steps, due to the integer representation different from the Cartesian coordinate system, the converted polar coordinate information will have many decimal angles and distances, which is not conducive to efficient calculation. Therefore, the embodiment of the present disclosure needs to perform non-uniform integer quantization processing on these angles and distances in combination with the internal characteristics of the rotary kiln. Specifically, the embodiment of the present disclosure can perform non-uniform integer quantization processing based on the size of the Cartesian coordinate information at each position in the feature image.

[0154] During the non-uniform integer quantization process, considering the circular characteristics of the rotary kiln, the area close to the axis will contain more details and important feature information, while the area far from the axis contains fewer details. Therefore, the disclosed embodiment can perform different levels of quantization on each Cartesian coordinate information based on the distance between each Cartesian coordinate information and the axis to retain as much important information as possible, so that in the area close to the axis, the quantization step size is smaller, that is, there are more quantization levels, and in the area far from the axis, the quantization step size is larger, that is, there are fewer quantization levels. Finally, the circular position coding information at each position in the feature image can be obtained based on the processed polar coordinate information, thereby reducing unnecessary calculation and storage requirements while retaining important details.

[0155] See also Fig.13 The embodiment of the present disclosure further provides a working condition identification device of a rotary kiln, which can implement the working condition identification method of the rotary kiln. The working condition identification device of the rotary kiln includes:

[0156] The image acquisition module 1301 is used to acquire a combustion image taken from the fire detection port of the rotary kiln toward the inside of the rotary kiln during the combustion process of the rotary kiln, and input the combustion image into a preset image classification model;

[0157] The circular position coding module 1302 is used to extract the characteristic image of the combustion image, and to perform circular position coding on the characteristic image with the position corresponding to the axis of the rotary kiln in the characteristic image as the center of the circle, so as to obtain the circular position coding information at each position in the characteristic image;

[0158] The coding embedding module 1303 is used to divide the feature image into multiple image blocks, and add the coding information of each circular position to the image block at the corresponding position to adjust the weight of each image block;

[0159] The working condition recognition module 1304 is used to perform self-attention processing on multiple image blocks based on the adjusted weights to obtain corresponding self-attention features, and perform image classification based on the self-attention features to obtain working condition recognition results during the rotary kiln combustion process.

[0160] In summary, the working condition identification device of the rotary kiln can first obtain the combustion image taken from the fire detection port of the rotary kiln to the inside of the rotary kiln during the combustion process of the rotary kiln by executing the working condition identification method of the rotary kiln, and input the combustion image into the preset image classification model. Since the rotary kiln has the physical characteristics of a rotating circle, the combustion image obtained from the fire detection port is a partial image of the circular cross-section of the rotary kiln. Therefore, after extracting the feature image of the combustion image, the position corresponding to the axis of the rotary kiln in the feature image is taken as the center of the circle, and the feature image is circularly encoded to obtain the circular position encoding information at each position in the feature image. Then, the feature image is divided into multiple image blocks, and each circular position encoding information is added to the image block at the corresponding position, so that the weight of each image block can be adjusted, and the position information expression ability of the material inside the rotary kiln body can be enhanced, and the feature can be better extracted. Finally, based on the adjusted weights, the multiple image blocks are self-attention processed to obtain the corresponding self-attention features, and the image is classified based on the self-attention features, so that the working condition identification result of the rotary kiln combustion process can be obtained, thereby improving the accuracy of the rotary kiln working condition identification.

[0161] The specific implementation of the rotary kiln working condition identification device is basically the same as the specific implementation of the rotary kiln working condition identification method described above, and will not be repeated here. On the premise of meeting the requirements of the embodiments of the present disclosure, the rotary kiln working condition identification device can also be provided with other functional modules to implement the rotary kiln working condition identification method described above.

[0162] The disclosed embodiment also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned rotary kiln working condition identification method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0163] See also Fig.14 , Fig.14 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0164] The processor 1401 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present disclosure;

[0165] The memory 1402 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1402 can store operating devices and other applications. When the technical solution provided in the embodiments of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1402, and the processor 1401 calls and executes the working condition identification method of the rotary kiln in the embodiment of the present disclosure;

[0166] Input / output interface 1403, used to implement information input and output;

[0167] The communication interface 1404 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.);

[0168] A bus 1405 that transmits information between various components of the device (e.g., the processor 1401, the memory 1402, the input / output interface 1403, and the communication interface 1404);

[0169] The processor 1401 , the memory 1402 , the input / output interface 1403 and the communication interface 1404 are connected to each other in communication within the device via a bus 1405 .

[0170] The embodiment of the present disclosure further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the operating condition identification method of the rotary kiln is implemented.

[0171] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. 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 disposed 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 combinations thereof.

[0172] The embodiments described in the embodiments of the present disclosure are intended to more clearly illustrate the technical solutions of the embodiments of the present disclosure and do not constitute a limitation on the technical solutions provided by the embodiments of the present disclosure. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present disclosure are also applicable to similar technical problems.

[0173] Those skilled in the art will appreciate that the technical solutions shown in the figures do not limit the embodiments of the present disclosure and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0174] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0175] Those skilled in the art will appreciate that all or some of the steps, devices, and functional modules / units in the above disclosed methods may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0176] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0177] It should be understood that in the present disclosure, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0178] In the several embodiments provided in the present disclosure, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0179] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0180] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0181] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present disclosure is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method of each embodiment of the present disclosure. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0182] The preferred embodiments of the present disclosure are described above with reference to the accompanying drawings, but the scope of the rights of the present disclosure is not limited thereto. Any modification, equivalent substitution and improvement made by those skilled in the art without departing from the scope and essence of the present disclosure should be within the scope of the rights of the present disclosure.

Claims

1. A method for identifying the working condition of a rotary kiln, characterized in that: include: Acquire a combustion image captured from a fire detection port of the rotary kiln toward the interior of the rotary kiln during the combustion process of the rotary kiln, and input the combustion image into a preset image classification model; Extracting a characteristic image of the combustion image, and taking a position in the characteristic image corresponding to the axis of the rotary kiln as the center of a circle, performing circular position coding on the characteristic image to obtain circular position coding information at each position in the characteristic image; Performing convolution processing on the feature image to adjust the number of channels to obtain a first convolution image; expanding the first convolution image to divide it into a plurality of image blocks, and adding each circular position encoding information to the image block at a corresponding position to adjust the weight of each image block; Based on the adjusted weights, self-attention processing is performed on the multiple image blocks to obtain corresponding self-attention features, and each of the self-attention features is folded to obtain a self-attention image, and the self-attention image is convolved to restore the number of channels to obtain a second convolution image; the second convolution image is spliced ​​with the feature image to obtain a spliced ​​image, and the spliced ​​image is convolved to restore the number of channels to obtain a third convolution image; image classification is performed based on the third convolution image to obtain an operating condition recognition result during the combustion process of the rotary kiln.

2. The method for identifying the working condition of a rotary kiln according to claim 1, characterized in that: The image classification model includes a plurality of mobile visual transformation blocks at different stages, the feature image is used as input data of any one of the mobile visual transformation blocks, and the third convolution image is output data of the corresponding mobile visual transformation block; The image classification is performed based on the third convolution image to obtain the working condition recognition result of the rotary kiln combustion process, including: The third convolution image is used as a feature image of the next stage, and is input into the moving visual transformation block of the next stage, until the moving visual transformation block of the last stage outputs the corresponding third convolution image, wherein the moving visual transformation blocks at multiple different stages share the same circular position code; Global pooling and full connection processing are performed based on the last third convolution image to obtain a fully connected feature, and image classification is performed based on the fully connected feature to obtain an operating condition recognition result during the combustion process of the rotary kiln.

3. The method for identifying the working condition of a rotary kiln according to claim 1, characterized in that: The image classification model is trained by the following steps, including: Acquire a plurality of sample images taken from a fire detection port of the rotary kiln toward the interior of the rotary kiln during the combustion process of the rotary kiln; Marking the working condition of each of the sample images to obtain a working condition label; Inputting each of the sample images into an initial image classification model to obtain each sample working condition recognition result; The initial image classification model is trained based on the sample working condition recognition results and the corresponding working condition labels until a training stop condition is met, thereby obtaining the trained image classification model.

4. The method for identifying the working condition of a rotary kiln according to claim 3, characterized in that: The method of obtaining a plurality of sample images captured from a fire detection port of the rotary kiln toward the interior of the rotary kiln during the combustion process of the rotary kiln includes: Acquire a sample video captured from a fire detection port of the rotary kiln toward the interior of the rotary kiln during the combustion process of the rotary kiln; The sample video is sliced ​​to obtain a plurality of video segments, and a video frame with a preset duration is selected from each of the video segments as a sample image.

5. The method for identifying the working condition of a rotary kiln according to claim 3 or 4, characterized in that: The step of marking the working condition of each sample image to obtain a working condition label includes: Acquire the working data of the rotary kiln corresponding to each of the sample images, wherein the working data includes at least one of the temperature at the head of the rotary kiln, the temperature at the tail of the rotary kiln, the frequency of the disc feeder, the frequency of the main motor, and the outlet pressure of the fan; The corresponding working condition of the rotary kiln is determined based on each of the working data, and the corresponding sample images are marked based on each of the working conditions to obtain a working condition label.

6. The method for identifying the working condition of a rotary kiln according to claim 1, characterized in that: The method of performing circular position coding on the characteristic image with the position corresponding to the axis of the rotary kiln in the characteristic image as the center of the circle to obtain circular position coding information at each position in the characteristic image includes: Taking the position in the characteristic image corresponding to the axis of the rotary kiln as the center of the circle, establishing Cartesian coordinate information of each position in the characteristic image; The Cartesian coordinate information at each position in the feature image is converted into polar coordinate information, and circular position coding information at each position in the feature image is obtained based on the polar coordinate information.

7. The method for identifying the working condition of a rotary kiln according to claim 6, characterized in that: The obtaining circular position coding information at each position in the feature image based on the polar coordinate information includes: Performing non-uniform integer quantization processing based on the size of the Cartesian coordinate information at each position in the feature image to obtain processed polar coordinate information; The circular position coding information at each position in the feature image is obtained based on the processed polar coordinate information.

8. A rotary kiln operating condition identification device, characterized in that: include: An image acquisition module is used to acquire a combustion image taken from a fire detection port of the rotary kiln toward the inside of the rotary kiln during the combustion process of the rotary kiln, and input the combustion image into a preset image classification model; A circular position coding module is used to extract the characteristic image of the combustion image, and to perform circular position coding on the characteristic image with the position corresponding to the axis of the rotary kiln in the characteristic image as the center of the circle, so as to obtain circular position coding information at each position in the characteristic image; A coding embedding module, configured to perform convolution processing on the feature image to adjust the number of channels to obtain a first convolution image; expand the first convolution image to divide it into a plurality of image blocks, and add the circular position coding information to the image block at the corresponding position to adjust the weight of each image block; The operating condition recognition module is used to perform self-attention processing on the multiple image blocks based on the adjusted weights to obtain corresponding self-attention features, and fold the various self-attention features to obtain a self-attention image, and perform convolution processing on the self-attention image to restore the number of channels to obtain a second convolution image; splicing the second convolution image with the feature image to obtain a spliced ​​image, and performing convolution processing on the spliced ​​image to restore the number of channels to obtain a third convolution image; performing image classification based on the third convolution image to obtain an operating condition recognition result during the combustion process of the rotary kiln.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method for identifying the working condition of a rotary kiln according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for identifying the working condition of a rotary kiln according to any one of claims 1 to 7 is implemented.

Images