Plasma stove abnormity monitoring method, device, equipment and medium

Abstract

The invention discloses a plasma stove abnormity monitoring method, device and equipment and a medium, and relates to the technical field of plasmas. According to the scheme, a first preset model and a second preset model are adopted to process and analyze image data and sensor data respectively, and the two models both adopt a deep learning algorithm; the accuracy and the reliability of subsequent abnormity judgment are improved; besides, by continuously monitoring the target state data corresponding to the plasma stove, the finally generated abnormal monitoring result of the plasma stove has real-time performance, and the safety of using the plasma stove is improved; meanwhile, along with continuous increase of new data, the first preset model and the second preset model can continuously learn and update so as to adapt to different working conditions and abnormal conditions, and the accuracy of generating the abnormal monitoring result of the plasma stove is further improved; in addition, the plasma stove abnormity monitoring method can be integrated and expanded with other safety monitoring systems to form a more perfect safety monitoring system.

Description

Technical Field

[0001] The present application relates to the field of plasma technology, and in particular to a method, device, equipment and medium for monitoring abnormalities in a plasma range. Background Art

[0002] With the rapid development of technology, kitchen appliances are undergoing rapid upgrading. Among them, plasma stoves, as an innovative cooking device, have won wide attention due to their high efficiency, environmental protection and energy saving advantages. However, this advanced equipment is also accompanied by a series of safety hazards during use, such as abnormal flame state and over-high temperature of plasma stoves. If these problems are not detected and handled properly, they are likely to trigger serious safety accidents.

[0003] At present, traditional plasma stove safety monitoring methods mainly rely on basic sensor equipment and manual judgment, but they all have problems such as delayed response, false alarms and missed alarms.

[0004] Therefore, how to accurately monitor the abnormalities of the plasma stove is an urgent problem to be solved. Summary of the invention

[0005] The present application aims to at least solve the technical problems existing in the prior art. To this end, the first aspect of the present application proposes a plasma stove abnormality monitoring method, the method comprising: Acquire target state data corresponding to the plasma stove; wherein the target state data includes target environment data, target current data and target flame image data; Using a first preset model to perform feature extraction processing on target flame image data to generate a target flame image region feature map; Using a second preset model to perform feature extraction processing on the target environment data and the target current data to generate target environment features and target current features; Perform feature fusion processing and global information extraction processing on the target flame image area feature map, target environment features and target current features to generate global features; Based on the global features, the plasma stove abnormal monitoring results are generated.

[0006] In a possible implementation, the first preset model is used to perform feature extraction processing on the target flame image data to generate a target flame image region feature map, including: Receive target flame image data through an input layer of a first preset model and distribute it to a preset multi-scale residual module, perform multi-scale feature extraction through a plurality of preset adaptive convolution layers in the preset multi-scale residual module, and obtain a multi-scale feature map; A multi-scale feature map is obtained through a preset feature enhancement module in the first preset model, and the multi-scale feature map is enhanced through a first preset self-attention mechanism module in the preset feature enhancement module to generate a target flame image area feature map.

[0007] In a possible implementation, the second preset model is used to perform feature extraction processing on the target environment data and the target current data to generate target environment features and target current features, including: Performing feature extraction processing on the target environment data and the target current data through a preset one-dimensional convolution layer in the second preset model to generate local environment features and local current features; Performing data conversion processing on the local environmental features and the local current features through a preset Transformer module in the second preset model to generate environmental features in a preset format and current features in a preset format; wherein the preset format is determined based on a self-attention mechanism in the preset Transformer module; The environment characteristics in a preset format and the current characteristics in a preset format are processed by a preset multi-layer encoder in the second preset model to generate target environment characteristics and target current characteristics.

[0008] In a possible implementation, feature fusion processing and global information extraction processing are performed on the target flame image region feature map, the target environment feature, and the target current feature to generate a global feature, including: The target flame image region feature map, target environment feature and target current feature are fused using different fully connected layers to generate fused features. The fused features are processed for global information extraction through the second preset self-attention mechanism module to generate global features.

[0009] In a possible implementation, different fully connected layers are used to perform feature fusion processing on the target flame image region feature map, the target environment feature, and the target current feature to generate fusion features, including: Different fully connected layers are used to perform dimension conversion processing on the target flame image region feature map, target environment feature and target current feature, respectively, to generate a normalized flame image region feature map, normalized environment feature and normalized current feature; Acquire a first weight corresponding to a normalized flame image region feature map, a second weight corresponding to a normalized environmental feature, and a third weight corresponding to a normalized current feature; Based on the first weight, the second weight and the third weight, a weighted sum process is performed on the normalized flame image region feature map, the normalized environmental feature and the normalized current feature to generate a fusion feature.

[0010] In a possible implementation, performing global information extraction processing on the fused features through a second preset self-attention mechanism module to generate global features includes: Mapping the fused features into a query vector, a key vector, and a value vector through a second preset self-attention mechanism module; The target attention weight is calculated by taking the dot product of the query vector and the key vector; The value vector is weighted summed based on the target attention weight to generate a global feature.

[0011] In a possible implementation manner, obtaining target state data corresponding to the plasma stove includes: Acquiring initial flame image data, initial environment data and initial current data; Performing image preprocessing on the initial flame image data to generate target flame image data; The initial environment data and the initial current data are respectively segmented and normalized according to a preset time window to generate target environment data and target current data; Target state data is obtained based on the target flame image data, the target environment data and the target current data.

[0012] The second aspect of the present application provides a plasma stove abnormality monitoring device, which includes: An acquisition module is used to acquire target state data corresponding to the plasma stove; wherein the target state data includes target environment data, target current data and target flame image data; A first extraction module, used for performing feature extraction processing on target flame image data using a first preset model to generate a target flame image region feature map; A second extraction module is used to perform feature extraction processing on the target environment data and the target current data using a second preset model to generate target environment features and target current features; A processing module is used to perform feature fusion processing and global information extraction processing on the target flame image area feature map, the target environment feature and the target current feature to generate a global feature; The generation module is used to generate plasma stove abnormality monitoring results based on global features.

[0013] In a possible implementation, the first extraction module is specifically used to: Receive target flame image data through an input layer of a first preset model and distribute it to a preset multi-scale residual module, perform multi-scale feature extraction through a plurality of preset adaptive convolution layers in the preset multi-scale residual module, and obtain a multi-scale feature map; A multi-scale feature map is obtained through a preset feature enhancement module in the first preset model, and the multi-scale feature map is enhanced through a first preset self-attention mechanism module in the preset feature enhancement module to generate a target flame image area feature map.

[0014] In a possible implementation manner, the second extraction module is specifically used to: Performing feature extraction processing on the target environment data and the target current data through a preset one-dimensional convolution layer in the second preset model to generate local environment features and local current features; Performing data conversion processing on the local environmental features and the local current features through a preset Transformer module in the second preset model to generate environmental features in a preset format and current features in a preset format; wherein the preset format is determined based on a self-attention mechanism in the preset Transformer module; The environment characteristics in a preset format and the current characteristics in a preset format are processed by a preset multi-layer encoder in the second preset model to generate target environment characteristics and target current characteristics.

[0015] In a possible implementation manner, the processing module is specifically used to: The target flame image region feature map, target environment feature and target current feature are fused using different fully connected layers to generate fused features. The fused features are processed for global information extraction through the second preset self-attention mechanism module to generate global features.

[0016] In a possible implementation manner, the processing module is further used to: Different fully connected layers are used to perform dimension conversion processing on the target flame image region feature map, target environment feature and target current feature, respectively, to generate a normalized flame image region feature map, normalized environment feature and normalized current feature; Acquire a first weight corresponding to a normalized flame image region feature map, a second weight corresponding to a normalized environmental feature, and a third weight corresponding to a normalized current feature; Based on the first weight, the second weight and the third weight, a weighted sum process is performed on the normalized flame image region feature map, the normalized environmental feature and the normalized current feature to generate a fusion feature.

[0017] In a possible implementation manner, the processing module is further used to: Mapping the fused features into a query vector, a key vector, and a value vector through a second preset self-attention mechanism module; The target attention weight is calculated by taking the dot product of the query vector and the key vector; The value vector is weighted summed based on the target attention weight to generate a global feature.

[0018] In a possible implementation manner, the acquisition module is specifically used to: Acquiring initial flame image data, initial environment data and initial current data; Performing image preprocessing on the initial flame image data to generate target flame image data; The initial environment data and the initial current data are respectively segmented and normalized according to a preset time window to generate target environment data and target current data; Target state data is obtained based on the target flame image data, the target environment data and the target current data.

[0019] In a third aspect, the present invention provides a device for plasma heating, comprising: Controller; a memory for storing instructions executable by the controller; Wherein, the controller is configured to execute the instructions to implement any of the plasma stove abnormality monitoring methods described above.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium, which stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by a processor to implement any one of the above-mentioned plasma stove abnormality monitoring methods.

[0021] The embodiments of the present application have the following beneficial effects: The plasma stove abnormality monitoring method, device, equipment and medium provided in the embodiments of the present application include: acquiring target state data corresponding to the plasma stove, using a first preset model to perform feature extraction processing on target flame image data to generate a target flame image area feature map, using a second preset model to perform feature extraction processing on target environment data and target current data to generate target environment features and target current features, performing feature fusion processing and global information extraction processing on the target flame image area feature map, target environment features and target current features to generate global features, and generating plasma stove abnormality monitoring results based on the global features. This solution uses a first preset model and a second preset model to process and analyze image data and sensor data respectively. Both the first preset model and the second preset model use deep learning related algorithms to improve the accuracy and reliability of subsequent abnormality judgments. In addition, by continuously monitoring the target state data corresponding to the plasma stove, the plasma stove abnormality monitoring result finally generated is real-time, which improves the safety of using the plasma stove. At the same time, as new data continues to increase, the first preset model and the second preset model can continue to learn and update to adapt to different working conditions and abnormal situations, further improving the accuracy of generating plasma stove abnormality monitoring results. Moreover, the plasma stove abnormality monitoring method can also be integrated and expanded with other safety monitoring systems to form a more complete safety monitoring system, further improving the safety of using the plasma stove. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flowchart of a plasma stove abnormality monitoring method provided in an embodiment of the present application; Figure 2 A flowchart of steps for obtaining target status data provided by an embodiment of the present application; Figure 3 A flow chart of steps for generating a target flame image region feature map provided by an embodiment of the present application; Figure 4 A flowchart of the steps of generating target environment characteristics and target current characteristics provided in an embodiment of the present application; Figure 5 A flowchart of the steps of performing feature fusion processing provided in an embodiment of the present application; Figure 6 A flowchart of the steps for generating fusion features provided in an embodiment of the present application; Figure 7 A flowchart of steps for generating global features provided in an embodiment of the present application; Figure 8 This is a structural block diagram of a plasma stove abnormality monitoring device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0024] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, unless otherwise specified, "multiple" means two or more. In addition, the use of "based on" or "according to" means openness and inclusiveness, because the process, steps, calculations or other actions "based on" or "according to" one or more of the conditions or values ​​may be based on additional conditions or values ​​beyond the described values ​​in practice.

[0025] Figure 1 This is a flow chart of the steps of a plasma stove abnormality monitoring method provided in an embodiment of the present application. Figure 1 As shown, the method comprises the following steps: Step 102: Obtain target state data corresponding to the plasma stove.

[0026] The target state data is used to characterize the relevant data of the plasma stove in the working state, and the target state data may include target environment data, target current data and target flame image data. Figure 2 As shown, Figure 2 A flowchart of steps for obtaining target status data provided in an embodiment of the present application includes: Step 202: Acquire initial flame image data, initial environment data, and initial current data.

[0027] The initial flame image data may be a flame image of a plasma cooker acquired in real time by an image acquisition device, and optionally, the image acquisition device may be a camera. The initial environmental data may include, but are not limited to, temperature data of a plasma cooker surface and surrounding environment acquired by a temperature sensor. The initial current data may be current data acquired by a current sensor after monitoring current fluctuations.

[0028] Step 204: perform image preprocessing on the initial flame image data to generate target flame image data.

[0029] Among them, the initial flame image data is an RGB image, and the initial flame image data can be spatially transformed from the RGB image to the Hue-Saturation-Value (HSV) space, and grayscale conversion is performed to obtain a grayscale image, which focuses on the brightness information of the image, removes color interference, and improves the focus area. Finally, the above grayscale image can be normalized to obtain the target flame image data. In addition, when there are multiple initial flame image data, each initial flame image data can also be sorted into an image sequence in chronological order.

[0030] Step 206 : performing segmentation processing and data normalization processing on the initial environment data and the initial current data according to the preset time window to generate target environment data and target current data.

[0031] After obtaining the initial environment data and the initial current data, the initial environment data and the initial current data can be processed in segments according to the preset time window to form a time series input. In addition, normalization processing is required to reduce the numerical difference, and finally the target environment data and the target current data are obtained. In addition, the target environment data and the target current data can be aligned with the target flame image data to ensure the temporal consistency of the multimodal data.

[0032] Step 208: Obtain target state data based on the target flame image data, the target environment data, and the target current data.

[0033] Among them, after the target flame image data, the target environment data and the target current data are generated, these three data can be used as target state data for subsequent multi-modal input.

[0034] Step 104: Use the first preset model to perform feature extraction processing on the target flame image data to generate a target flame image region feature map.

[0035] After the target flame image data is acquired, the first preset model may be used to perform feature extraction processing on the target flame image data, thereby generating a target flame image region feature map.

[0036] In some optional embodiments, Figure 3 As shown, Figure 3 A flowchart of the steps of generating a target flame image region feature map provided in an embodiment of the present application includes: Step 302: Receive target flame image data through the input layer of the first preset model and distribute it to the preset multi-scale residual module, perform multi-scale feature extraction through multiple preset adaptive convolution layers in the preset multi-scale residual module, and obtain a multi-scale feature map.

[0037] Step 304: obtain a multi-scale feature map through the preset feature enhancement module in the first preset model, enhance the multi-scale feature map through the first preset self-attention mechanism module in the preset feature enhancement module, and generate a target flame image area feature map.

[0038] The first preset model is based on the residual network (Residual Network, referred to as ResNet), and adds a preset multi-scale residual module, a preset adaptive convolutional layer, and a preset feature enhancement module. ResNet is a classic neural network that serves as the backbone of many computer vision tasks.

[0039] Thus, the target flame image data can be received through the input layer of the first preset model and distributed to the preset multi-scale residual module, and multi-scale feature extraction is performed through multiple preset adaptive convolution layers in the preset multi-scale residual module to obtain a multi-scale feature map. Specifically, the input layer takes the target flame image data as input and sets 9 residual blocks. In addition, the convolution kernel size of the initial convolution layer is set to 7×7, the step size is set to 2, and the number of channels is 32. Using a larger convolution kernel can capture the edges, corners and simple textures of the image. The step size is set to 2, and the convolution operation will reduce the spatial size of the image by half, reducing the computational complexity, while increasing the number of channels to 32 to extract higher-dimensional features.

[0040] Traditional ResNet networks usually use convolutions of a single size to extract features, but in this application, a preset multi-scale residual module is added, which uses convolutions of multiple scales to capture features at different levels. Exemplarily, three convolution layers of different scales can be included: 3×3, 5×5, and 7×7 convolutions. The first convolution layer (3×3): mainly extracts local details, such as small-scale textures or edges. The second convolution layer (5×5): captures medium-scale features, which is suitable for the edges of the flame state and the middle part of the flame. The third convolution layer (7×7): a convolution with a larger receptive field, used to capture global features such as flame morphology and combustion stability.

[0041] In addition, in the above convolution operation, the adaptive convolution layer technology is used. By using a 1×1 convolution layer to generate a dynamic convolution kernel, the convolution kernel will be adjusted according to the different input features. The dynamic convolution kernel generates a richer feature representation by element-by-element convolution with the input feature map. Moreover, the convolution kernel can be dynamically adjusted according to the characteristics of the input target flame image data, so that the model can adaptively process the details of different target flame image data and enhance the network's perception of flames.

[0042] After each convolution layer, a batch normalization (BN) layer and a ReLU activation function can be added to perform weighted sum fusion on the feature maps obtained from different convolution kernels to obtain the final multi-scale feature map. Standardizing the output of the convolution operation through BN can accelerate model training and stabilize the learning process. The ReLU activation function introduces the nonlinear ability of the network through nonlinear transformation, enabling the model to learn complex features and relationships.

[0043] In order to further improve the feature extraction capability of the first preset model for the target flame image data, a first preset self-attention mechanism module can also be added after the preset multi-scale residual module. Thus, a multi-scale feature map can be obtained through the preset feature enhancement module in the first preset model, and the multi-scale feature map is enhanced by the first preset self-attention mechanism module in the preset feature enhancement module to generate a target flame image region feature map. Among them, the multi-scale feature map can be fused with the feature information in the initial flame image data, and optionally, the concatenation function concat can be used to combine the features after different processing, and finally generate a richer target flame image region feature map. The target flame image region feature map is a feature vector used to characterize the brightness, shape and dynamic characteristics of the flame.

[0044] The above-mentioned first preset self-attention mechanism module enhances the expressiveness of the network by dynamically adjusting the importance of each channel. In each residual block, the global feature information of each channel is obtained by global average pooling. Then a small fully connected layer and ReLU activation function are used to generate a weight vector, which is used to adjust the weight of each channel. Finally, the weight vector and the input feature map are weighted channel by channel to highlight important channels and suppress unimportant channels. In addition, by learning the weight of each channel, the network can pay more attention to the channels that play an important role in plasma flame image detection during the training process, thereby enhancing the key features of the plasma flame image such as color and morphology, and improving the classification and recognition accuracy of the model.

[0045] Step 106: Use the second preset model to perform feature extraction processing on the target environment data and the target current data to generate target environment features and target current features.

[0046] Among them, the second preset model is a model that combines a preset one-dimensional convolutional layer with a preset Transformer module. Therefore, after obtaining the target environment data and the target current data, the second preset model can be used to perform feature extraction processing on the target environment data and the target current data to generate target environment features and target current features. The target environment features and target current features are respectively used to characterize feature vectors representing temperature changes and current changes.

[0047] In some optional embodiments, Figure 4 As shown, Figure 4 A flowchart of steps for generating target environment characteristics and target current characteristics provided in an embodiment of the present application includes: Step 402: Perform feature extraction processing on the target environment data and the target current data through a preset one-dimensional convolution layer in the second preset model to generate local environment features and local current features.

[0048] Step 404: Perform data conversion processing on the local environmental characteristics and the local current characteristics through the preset Transformer module in the second preset model to generate environmental characteristics in a preset format and current characteristics in a preset format.

[0049] Step 406: Process the environment characteristics in a preset format and the current characteristics in a preset format through a preset multi-layer encoder in the second preset model to generate target environment characteristics and target current characteristics.

[0050] Among them, a preset one-dimensional convolution layer is used as a feature extraction module to convert the target environment data and target current data as time series data into feature vectors. Since the changes in temperature and current data often manifest as local fluctuations (such as excessive temperature, sudden current changes, etc.), a preset one-dimensional convolution layer is used to capture the local time series characteristics of the data and automatically extract local patterns in the data, such as sudden temperature rise or current fluctuations, so that local environmental features and local current features can be extracted through the preset one-dimensional convolution layer.

[0051] Next, the local environmental features and local current features can be converted and processed by the preset Transformer module to generate environmental features in a preset format and current features in a preset format, wherein the preset format is determined based on the self-attention mechanism in the preset Transformer module. Specifically, the preset Transformer module is used to capture global dependencies in time series data, that is, to capture long-term trends or periodic changes in time series. The preset Transformer module calculates the associations between different time steps through the self-attention mechanism, and can effectively handle long-term dependencies. After the input passes through the preset one-dimensional convolutional layer to extract local features, the preset Transformer module converts the data into a preset format suitable for the self-attention mechanism, and calculates the associations between each time step.

[0052] The environmental features in a preset format and the current features in a preset format are then processed by a preset multi-layer encoder to gradually capture the global information of the time series, and output a representation containing global time-dependent features, thereby capturing the long-term pattern in the sequence.

[0053] Finally, through the processing of the preset one-dimensional convolutional layer and the preset Transformer module, the model outputs a high-dimensional feature layer, which represents the key patterns and features in the temperature and current time series, that is, the target environment characteristics and target current characteristics are obtained.

[0054] Step 108: performing feature fusion processing and global information extraction processing on the target flame image region feature map, the target environment feature and the target current feature to generate a global feature.

[0055] Among them, after obtaining the target flame image area feature map, target environment features and target current features, the multimodal features can be subjected to feature fusion processing and global information extraction processing to finally generate global features.

[0056] In some optional embodiments, Figure 5 As shown, Figure 5 A flowchart of the steps of performing feature fusion processing provided in an embodiment of the present application includes: Step 502: perform feature fusion processing on the target flame image region feature map, the target environment feature and the target current feature using different fully connected layers to generate fusion features.

[0057] Wherein, when performing feature fusion processing, optionally, the target flame image region feature map, the target environment feature and the target current feature can be directly spliced ​​to obtain the fusion feature.

[0058] In some other optional embodiments, Figure 6 As shown, Figure 6 A flowchart of steps for generating fusion features provided in an embodiment of the present application includes: Step 602: Perform dimensionality conversion processing on the target flame image region feature map, the target environment feature and the target current feature using different fully connected layers to generate a normalized flame image region feature map, a normalized environment feature and a normalized current feature.

[0059] Step 604: Obtain a first weight corresponding to the normalized flame image region feature map, a second weight corresponding to the normalized environmental feature, and a third weight corresponding to the normalized current feature.

[0060] Step 606: Perform weighted sum processing on the normalized flame image region feature map, the normalized environmental feature, and the normalized current feature based on the first weight, the second weight, and the third weight to generate a fusion feature.

[0061] Among them, since the feature dimensions of different modes may be different, it is necessary to align the features of each mode first, and then use different fully connected layers to perform dimension conversion processing on the target flame image region feature map, target environment features and target current features respectively to generate normalized flame image region feature map, normalized environment features and normalized current features. The features of each mode are dimensionally converted through an independent fully connected layer, so that the features of all modes are unified to the same feature dimension. In this process, the nonlinear activation function ReLU can be used to enhance the feature expression ability and generate a normalized feature representation.

[0062] In addition, corresponding weights can be assigned to the features of each modality in advance, and these weights can be obtained through network learning, reflecting the importance of the features of different modalities in the task. Thus, a first weight corresponding to the normalized flame image region feature map, a second weight corresponding to the normalized environmental feature, and a third weight corresponding to the normalized current feature can be obtained. Finally, based on the first weight, the second weight, and the third weight, the normalized flame image region feature map, the normalized environmental feature, and the normalized current feature can be weighted and summed to generate a fusion feature.

[0063] Step 504: Perform global information extraction processing on the fused features through the second preset self-attention mechanism module to generate global features.

[0064] Among them, the global information of the fused features can be further extracted through the second preset self-attention mechanism module to capture the mutual relationship between modalities and the dynamic characteristics of the time series.

[0065] In some optional embodiments, Figure 7 As shown, Figure 7 A flowchart of steps for generating global features provided in an embodiment of the present application includes: Step 702: Map the fused features into a query vector, a key vector, and a value vector through a second preset self-attention mechanism module.

[0066] Step 704: Calculate the target attention weight by taking the dot product of the query vector and the key vector.

[0067] Step 706: Perform weighted summation on the value vector based on the target attention weight to generate a global feature.

[0068] Among them, the fused features can be mapped into query vectors, key vectors and value vectors by using different linear changes through the second preset self-attention mechanism module, and then the target attention weight is calculated by the dot product of the query vector and the key vector. Finally, the value vector can be weighted and summed based on the target attention weight to generate global features.

[0069] Step 110: Generate plasma stove abnormality monitoring results based on global features.

[0070] Among them, the global features can be processed through a series of fully connected layers to further explore high-level feature expressions. Finally, the classifier can use the Sigmoid activation function to output the prediction results of anomaly detection to determine whether the system is in an abnormal state, thereby generating plasma stove abnormal monitoring results.

[0071] Optionally, the prediction results may be decoded, the degree of abnormality may be determined according to the size of the corresponding scores, and based on a preset alarm threshold, the scores exceeding the alarm threshold may be marked as abnormal.

[0072] Furthermore, the prediction results can be transmitted to the alarm system or display interface. If an abnormality is detected, an alarm is triggered or safety measures such as shutting down the plasma stove are taken. In this way, the working status of the plasma stove can be monitored in real time, and an early warning can be issued immediately when an abnormality occurs, thereby improving safety.

[0073] In some optional embodiments, the first preset model and the second preset model need to be pre-trained. Specifically, in the process of data set preparation, data enhancement can be performed on the environmental data set and the current data set respectively, which can include adding Gaussian noise to simulate the error in sensor measurement or the noise caused by environmental changes, and enhancing the robustness of the model to imperfect data; and performing time shifting to simulate changes at different time points. In order to adapt to different time series modes, for example, the data of a certain period of time can be randomly moved forward or backward by a certain time window, and then the corresponding labels can be re-labeled.

[0074] In addition, data enhancement is also required for the flame image dataset, which can include random rotation from -45° to 45° and flipping to simulate the flame state taken from different perspectives; reducing the brightness and adding salt and pepper noise to simulate the flame taken in a night environment; performing Gaussian blur to simulate the camera out of focus; and performing random area stitching, that is, covering part of the area of ​​another flame image on the original image.

[0075] According to the data enhancement method proposed above, the environmental data set, the current data set and the flame image data set are enhanced, and the data are manually labeled to determine whether they are abnormal data. Each data set is divided into a training set, a validation set and a test set according to the proportion. For example, the ratio of the training set, the validation set and the test set can be 70%:15%:15%.

[0076] Next, the He initialization method can be used to initialize the network parameters of the first initial model and the network parameters of the second initial model, and the Adam optimization algorithm is selected, and the initial learning rate is set to 1e -3, the loss function uses cross entropy loss. He initialization is a robust neural network parameter initialization method proposed by He Kaiming et al., which can ensure that information can flow effectively during forward propagation and back propagation, making the variance of input signals in different layers roughly equal.

[0077] When training the model, the flame image training set can be input into the first initial model, and the environment training set and the current training set can be input into the second initial model as time series. After the features of each modality are extracted by the corresponding model, feature fusion processing and global information extraction processing are performed to finally generate global features. The fused global features are processed by the attention mechanism and finally enter the fully connected layer to output the prediction results. The loss value is calculated based on the prediction results output by the model and the manually annotated true labels. The loss value is used to measure the performance of the current model, and the gradient is calculated using the back propagation algorithm.

[0078] The present application provides a plasma stove abnormality monitoring method, which includes: acquiring target state data corresponding to the plasma stove, using a first preset model to perform feature extraction processing on target flame image data to generate a target flame image area feature map, using a second preset model to perform feature extraction processing on target environment data and target current data to generate target environment features and target current features, performing feature fusion processing and global information extraction processing on the target flame image area feature map, target environment features and target current features to generate global features, and generating plasma stove abnormality monitoring results based on the global features. This solution uses a first preset model and a second preset model to process and analyze image data and sensor data respectively. Both the first preset model and the second preset model use deep learning related algorithms to improve the accuracy and reliability of subsequent abnormality judgments. In addition, by continuously monitoring the target state data corresponding to the plasma stove, the plasma stove abnormality monitoring result finally generated is real-time, which improves the safety of using the plasma stove. At the same time, as new data continues to increase, the first preset model and the second preset model can continue to learn and update to adapt to different working conditions and abnormal situations, further improving the accuracy of generating plasma stove abnormality monitoring results. Moreover, the plasma stove abnormality monitoring method can also be integrated and expanded with other safety monitoring systems to form a more complete safety monitoring system, further improving the safety of using the plasma stove.

[0079] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0080] Figure 8 This is a structural block diagram of a plasma stove abnormality monitoring device provided in an embodiment of the present application.

[0081] like Figure 8 As shown, the plasma stove abnormality monitoring device 800 includes: The acquisition module 802 is used to acquire target state data corresponding to the plasma stove; wherein the target state data includes target environment data, target current data and target flame image data.

[0082] The first extraction module 804 is used to perform feature extraction processing on the target flame image data using a first preset model to generate a target flame image region feature map.

[0083] The second extraction module 806 is used to perform feature extraction processing on the target environment data and the target current data using a second preset model to generate target environment features and target current features.

[0084] The processing module 808 is used to perform feature fusion processing and global information extraction processing on the target flame image area feature map, the target environment feature and the target current feature to generate a global feature.

[0085] The generation module 810 is used to generate plasma stove abnormality monitoring results based on global features.

[0086] Regarding the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here. Each module in the above-mentioned graph-based retrieval enhancement generation device can be implemented in whole or in part by software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations of the above modules.

[0087] In one embodiment of the present application, there is provided a device for plasma heating, comprising: Controller; a memory for storing instructions executable by the controller; Wherein, the controller is configured to execute the instructions to implement any of the plasma stove abnormality monitoring methods described above.

[0088] The device using plasma heating provided in the embodiment of the present application has similar implementation principles and technical effects as those of the above-mentioned method embodiment, and will not be described in detail here.

[0089] In one embodiment of the present application, an embodiment of the present invention also provides a computer-readable storage medium, in which at least one instruction, at least one program, code set or instruction set is stored, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the plasma stove abnormality monitoring method in the embodiment of the present invention.

[0090] The computer-readable storage medium provided in this embodiment has similar implementation principles and technical effects to those of the above method embodiments, and will not be described in detail here.

[0091] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the process or function according to the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be accessed by the computer or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0092] It is easy to understand that those skilled in the art can combine, split, reorganize, etc. the embodiments of the present application to obtain other embodiments based on the several embodiments provided in the present application, and these embodiments do not exceed the protection scope of the present application.

[0093] The above specific implementation methods further explain in detail the purpose, technical solutions and beneficial effects of the embodiments of the present application. It should be understood that the above are only specific implementation methods of the embodiments of the present application and are not used to limit the protection scope of the embodiments of the present application. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solutions of the embodiments of the present application should be included in the protection scope of the embodiments of the present application.

Claims

1. A plasma stove abnormality monitoring method, characterized in that: The method comprises: Acquire target state data corresponding to the plasma stove; wherein the target state data includes target environment data, target current data and target flame image data; Using a first preset model to perform feature extraction processing on the target flame image data to generate a target flame image region feature map; Using a second preset model to perform feature extraction processing on the target environment data and the target current data to generate target environment features and target current features; Performing feature fusion processing and global information extraction processing on the target flame image area feature map, the target environment feature and the target current feature to generate a global feature; Based on the global features, plasma stove abnormality monitoring results are generated.

2. The method according to claim 1, characterized in that The method of using the first preset model to perform feature extraction processing on the target flame image data to generate a target flame image region feature map includes: The target flame image data is received through the input layer of the first preset model and distributed to a preset multi-scale residual module, and multi-scale feature extraction is performed through a plurality of preset adaptive convolution layers in the preset multi-scale residual module to obtain a multi-scale feature map; The multi-scale feature map is obtained through the preset feature enhancement module in the first preset model, and the multi-scale feature map is enhanced through the first preset self-attention mechanism module in the preset feature enhancement module to generate the target flame image area feature map.

3. The method according to claim 1 or 2, characterized in that: The adopting the second preset model to perform feature extraction processing on the target environment data and the target current data to generate target environment features and target current features includes: Performing feature extraction processing on the target environment data and the target current data through a preset one-dimensional convolution layer in the second preset model to generate local environment features and local current features; Performing data conversion processing on the local environmental features and the local current features through a preset Transformer module in the second preset model to generate environmental features in a preset format and current features in a preset format; wherein the preset format is determined based on a self-attention mechanism in the preset Transformer module; The environment feature in the preset format and the current feature in the preset format are processed by a preset multi-layer encoder in the second preset model to generate the target environment feature and the target current feature.

4. The method according to claim 1 or 2, characterized in that: The step of performing feature fusion processing and global information extraction processing on the target flame image region feature map, the target environment feature and the target current feature to generate a global feature includes: Using different fully connected layers to perform feature fusion processing on the target flame image region feature map, the target environment feature and the target current feature, respectively, to generate fusion features; The fused features are processed for global information extraction through a second preset self-attention mechanism module to generate the global features.

5. The method according to claim 4, characterized in that The step of using different fully connected layers to perform feature fusion processing on the target flame image region feature map, the target environment feature and the target current feature to generate fusion features includes: The target flame image region feature map, the target environment feature and the target current feature are respectively subjected to dimension conversion processing using different fully connected layers to generate a normalized flame image region feature map, a normalized environment feature and a normalized current feature; Acquire a first weight corresponding to the normalized flame image region feature map, a second weight corresponding to the normalized environmental feature, and a third weight corresponding to the normalized current feature; The normalized flame image region feature map, the normalized environmental feature and the normalized current feature are weighted-summed based on the first weight, the second weight and the third weight to generate the fusion feature.

6. The method according to claim 4, characterized in that The step of performing global information extraction processing on the fusion feature through the second preset self-attention mechanism module to generate the global feature includes: Mapping the fused features into a query vector, a key vector and a value vector through the second preset self-attention mechanism module; Calculate the target attention weight by taking the dot product of the query vector and the key vector; The value vector is weightedly summed based on the target attention weight to generate the global feature.

7. The method according to claim 1 or 2, characterized in that: The step of obtaining target state data corresponding to the plasma stove includes: Acquiring initial flame image data, initial environment data and initial current data; Performing image preprocessing on the initial flame image data to generate the target flame image data; Performing segmentation processing and data normalization processing on the initial environment data and the initial current data according to a preset time window to generate the target environment data and the target current data; The target state data is obtained based on the target flame image data, the target environment data and the target current data.

8. A plasma stove abnormality monitoring device, characterized in that: The device comprises: An acquisition module, used for acquiring target state data corresponding to the plasma stove; wherein the target state data includes target environment data, target current data and target flame image data; A first extraction module, used for performing feature extraction processing on the target flame image data using a first preset model to generate a target flame image region feature map; A second extraction module, used for performing feature extraction processing on the target environment data and the target current data using a second preset model to generate target environment features and target current features; A processing module, used for performing feature fusion processing and global information extraction processing on the target flame image area feature map, the target environment feature and the target current feature to generate a global feature; A generation module is used to generate plasma stove abnormality monitoring results based on the global features.

9. A device using plasma heating, characterized in that: include: Controller; a memory for storing instructions executable by the controller; Wherein, the controller is configured to execute the instructions to implement the plasma stove abnormality monitoring method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the plasma stove abnormality monitoring method as described in any one of claims 1-7.

Images