Power image quality evaluation method and system based on adaptive super-network architecture

Abstract

The invention provides a power image quality evaluation method and system based on an adaptive super network architecture, and the method comprises the steps: inputting an obtained to-be-evaluated power image into a pre-constructed image semantic feature extraction network, and extracting local features and global features of multiple levels; based on the global features, generating global image quality discrimination parameters by using an adaptive super network; inputting the local features of the plurality of levels into a target prediction network constructed based on global image quality discrimination parameters to obtain an image quality score of the to-be-evaluated power image; the local features and the global features are extracted through the feature extraction network, subsequent evaluation is performed on the image based on the extracted features, the problem of difficult evaluation index construction is solved, then the network parameters in the target prediction network are generated by using the adaptive super network based on the global features, and the image is evaluated through the target prediction network. The problem of evaluation rule solidification is solved, and generalization and accuracy of model evaluation are improved.

Description

Technical Field

[0001] The present invention belongs to the field of image processing, and in particular relates to a method and system for evaluating power image quality based on an adaptive super network architecture. Background Art

[0002] With the rapid development of the power industry, the scale of the power system continues to expand, and the safety and reliability of the power system operation are becoming increasingly important. Power image evaluation and analysis technology can efficiently and accurately process and analyze image data in the power system in an automated way, thereby improving the operating efficiency and safety of the power system. At present, the analysis of power images is mainly based on traditional machine learning. First, various indicators such as resolution, average gradient, average brightness, etc. are constructed through image processing technology. Then, on the manually annotated data set, support vector machines, random forests and other machine learning algorithms are used to iteratively train on the data set to learn the relationship between each indicator and the image quality score. Through the trained model, the image quality of the newly input image can be evaluated.

[0003] However, the problems with image quality evaluation methods based on traditional machine learning are: it is necessary to manually construct evaluation indicators related to image quality. For different scenarios and different image quality problems, the selection of indicators is also different. Therefore, the rationality of the construction of evaluation indicators will directly affect the accuracy of image quality evaluation. In addition, the traditional method uses machine learning algorithms to train the model. For any input image, the same parameters are used for quality evaluation. The evaluation rules are rigid and cannot adapt to scenarios with relatively complex business content. The image evaluation model has poor generalization and low accuracy. Summary of the invention

[0004] In order to overcome the above-mentioned deficiencies of the prior art, the present invention proposes a power image quality assessment method based on an adaptive super network architecture, comprising:

[0005] The acquired power image to be evaluated is input into a pre-built image semantic feature extraction network to extract local features and global features at multiple levels;

[0006] Based on the global features, using an adaptive super network to generate global image quality discrimination parameters;

[0007] The local features of the multiple levels are input into a target prediction network constructed based on the global image quality discrimination parameter to obtain an image quality score of the power image to be evaluated.

[0008] Preferably, the acquired power image to be evaluated is input into a pre-built image semantic feature extraction network to extract local features and global features at multiple levels, including:

[0009] Input the acquired power image to be evaluated into a pre-built image semantic feature extraction network;

[0010] Through multiple convolutional stacking modules connected in series in the image semantic feature extraction network, local features of multiple levels are extracted in sequence, and the local features of the highest level extracted by the last convolutional stacking module are used as global features.

[0011] Preferably, the global image quality discrimination parameters include: local feature global evaluation parameters of each level and global feature evaluation parameters;

[0012] The method of generating global image quality discrimination parameters based on the global features by using an adaptive super network includes:

[0013] Building an adaptive super network based on the total number of levels of the local features and the local feature size information of each level;

[0014] The global features are input into the adaptive super network to generate the local feature global evaluation parameters of each level and the global feature evaluation parameters.

[0015] Preferably, the target prediction network includes multiple levels of convolutional layers, multiple activation function layers, connection layers and linear mapping layers;

[0016] Using the inputs of the convolutional layers of the multiple levels as inputs of the target prediction network;

[0017] The convolution layer of each level is used to further learn and compress the local features of the corresponding level of the input based on the global feature information, obtain the local features of multiple levels after convolution, and output them to the corresponding activation function layer;

[0018] The activation function layer is used to map the local features after convolution and output them to the connection layer;

[0019] The connection layer is used to fuse the output results of multiple activation function layers, and then input the fused local features into the linear mapping layer for calculation to obtain an image quality score as the output of the target prediction network;

[0020] The weight parameters of each convolutional layer are determined based on the global evaluation parameters of the local features of each level, and the weight parameters of the linear mapping layer are determined based on the global feature evaluation parameters.

[0021] Preferably, the global feature evaluation parameters include: a global feature evaluation weight parameter and a global feature evaluation bias parameter; the expression of the image quality score is as follows:

[0022] Y=Lconcat W n +b n

[0023] Where: Y is the image quality score; W n is the weight parameter for global feature evaluation; b n Bias parameter for global feature evaluation; L concat is the fused local feature.

[0024] Preferably, the image semantic feature extraction network comprises: a preset number of convolution stacking modules and pooling layers, wherein the convolution stacking modules and corresponding pooling layers are stacked in sequence;

[0025] The input of the first convolution stack module is used as the input of the image semantic feature extraction network to extract and output the first layer of local features of the power image to be evaluated;

[0026] The output of the first convolution stack module is used as the input of the first pooling layer. After normalizing the local features of the first layer, it is input into the second convolution stack module to extract and output the local features of the second layer.

[0027] After the local features of the second layer pass through the next pooling layer, they are input into the next convolutional stacking module to extract and output the local features of the next layer, until the last convolutional stacking module extracts the local features of the highest level as global features, and then the local features and global features of multiple levels output by each convolutional stacking module are used as the output of the image semantic feature extraction network.

[0028] Preferably, the convolution stacking module includes a first 1x1 convolution layer, a second 1x1 convolution layer, a 3x3 convolution layer and a batch normalization layer connected in sequence, wherein the first 1x1 convolution layer is the input end of the convolution stacking module, the second 1x1 convolution layer is the output end of the convolution stacking module, the 3x3 convolution layer is used to extract local features, and the batch normalization layer is used to regularize the output of each convolution layer.

[0029] Based on the same inventive concept, the present invention also proposes a power image quality assessment system based on an adaptive super network architecture, comprising:

[0030] Feature extraction module: used to input the acquired power image to be evaluated into a pre-built image semantic feature extraction network to extract local features and global features at multiple levels;

[0031] Parameter generation module: used to generate global image quality discrimination parameters based on the global features using an adaptive super network;

[0032] Image quality assessment module: used to input the local features of the multiple levels into the target prediction network constructed based on the global image quality discrimination parameters to obtain the image quality score of the power image to be evaluated.

[0033] Preferably, the feature extraction module is specifically used for:

[0034] Input the acquired power image to be evaluated into a pre-built image semantic feature extraction network;

[0035] Through multiple convolutional stacking modules connected in series in the image semantic feature extraction network, local features of multiple levels are extracted in sequence, and the local features of the highest level extracted by the last convolutional stacking module are used as global features.

[0036] Preferably, the global image quality discrimination parameters include: local feature global evaluation parameters of each level and global feature evaluation parameters;

[0037] The parameter generation module is specifically used for:

[0038] Building an adaptive super network based on the total number of levels of the local features and the local feature size information of each level;

[0039] The global features are input into the adaptive super network to generate the local feature global evaluation parameters of each level and the global feature evaluation parameters.

[0040] Preferably, the target prediction network in the image quality assessment module includes multiple levels of convolutional layers, multiple activation function layers, connection layers and linear mapping layers;

[0041] Using the inputs of the convolutional layers of the multiple levels as inputs of the target prediction network;

[0042] The convolution layer of each level is used to further learn and compress the local features of the corresponding level of the input based on the global feature information, obtain the local features of multiple levels after convolution, and output them to the corresponding activation function layer;

[0043] The activation function layer is used to map the local features after convolution and output them to the connection layer;

[0044] The connection layer is used to fuse the output results of multiple activation function layers, and then input the fused local features into the linear mapping layer for calculation to obtain an image quality score as the output of the target prediction network;

[0045] The weight parameters of each convolutional layer are determined based on the global evaluation parameters of the local features of each level, and the weight parameters of the linear mapping layer are determined based on the global feature evaluation parameters.

[0046] Preferably, the global feature evaluation parameters in the image quality assessment module include: a global feature evaluation weight parameter and a global feature evaluation bias parameter; the expression of the image quality score is as follows:

[0047] Y=L concat W n +b n

[0048] Where: Y is the image quality score; W n is the weight parameter for global feature evaluation; b n Bias parameter for global feature evaluation; L concat is the fused local feature.

[0049] Preferably, the image semantic feature extraction network in the feature extraction module includes: a preset number of convolution stacking modules and pooling layers, wherein the convolution stacking modules and corresponding pooling layers are stacked in sequence;

[0050] The input of the first convolution stack module is used as the input of the image semantic feature extraction network to extract and output the first layer of local features of the power image to be evaluated;

[0051] The output of the first convolution stack module is used as the input of the first pooling layer. After normalizing the local features of the first layer, it is input into the second convolution stack module to extract and output the local features of the second layer.

[0052] After the local features of the second layer pass through the next pooling layer, they are input into the next convolutional stacking module to extract and output the local features of the next layer, until the last convolutional stacking module extracts the local features of the highest level as global features, and then the local features and global features of multiple levels output by each convolutional stacking module are used as the output of the image semantic feature extraction network.

[0053] Preferably, the convolution stacking module in the feature extraction module includes a first 1x1 convolution layer, a second 1x1 convolution layer, a 3x3 convolution layer and a batch normalization layer connected in sequence, wherein the first 1x1 convolution layer is the input end of the convolution stacking module, the second 1x1 convolution layer is the output end of the convolution stacking module, the 3x3 convolution layer is used to extract local features, and the batch normalization layer is used to regularize the output of each convolution layer.

[0054] Based on the same inventive concept, the present invention also proposes an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus;

[0055] The memory is used to store one or more programs;

[0056] When the one or more programs are executed by the at least one processor, a power image quality assessment method based on an adaptive super network architecture as described above is implemented.

[0057] Based on the same inventive concept, the present invention also proposes a readable storage medium having an execution program stored thereon. When the execution program is executed, a power image quality assessment method based on an adaptive super network architecture as described above is implemented.

[0058] Compared with the closest prior art, the present invention has the following beneficial effects:

[0059] The present invention provides a method and system for evaluating the quality of electric power images based on an adaptive super network architecture, comprising: inputting an acquired electric power image to be evaluated into a pre-constructed image semantic feature extraction network to extract local features and global features of multiple levels; based on the global features, using an adaptive super network to generate global image quality discrimination parameters; inputting the local features of the multiple levels into a target prediction network constructed based on the global image quality discrimination parameters to obtain an image quality score for the electric power image to be evaluated; the present invention extracts local features and global features through a feature extraction network, and subsequently evaluates the image based on the extracted features, thereby solving the problem of difficulty in constructing evaluation indicators; then, based on the global features, using an adaptive super network to generate network parameters in a target prediction network, and evaluating the image through the target prediction network, thereby solving the problem of solidified evaluation rules and improving the generalization and accuracy of model evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic flow chart of a method for evaluating power image quality based on an adaptive super network architecture provided by the present invention;

[0061] Figure 2 A structural schematic diagram of a convolution stacking module provided by the present invention;

[0062] Figure 3 A schematic diagram of a network structure for extracting semantic features of an image according to the present invention;

[0063] Figure 4 A schematic diagram of an adaptive super network structure provided by the present invention;

[0064] Figure 5 A schematic diagram of a target prediction network structure provided by the present invention;

[0065] Figure 6 A schematic diagram of a network structure of an overall image quality evaluation model provided by the present invention;

[0066] Figure 7A schematic diagram of the structure of a power image quality assessment system based on an adaptive super network architecture provided by the present invention;

[0067] Figure 8 A schematic diagram of an electronic device for power image quality assessment based on an adaptive super network architecture provided by the present invention. DETAILED DESCRIPTION

[0068] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings.

[0069] Embodiment 1:

[0070] The present invention provides a method for evaluating power image quality based on an adaptive super network architecture, specifically, Figure 1 As shown, it is a flow chart of a method for evaluating power image quality based on an adaptive super network architecture provided by the present invention, comprising the following steps:

[0071] Step 1: Input the acquired power image to be evaluated into the pre-built image semantic feature extraction network to extract local features and global features at multiple levels;

[0072] Step 2: Based on the global features, using an adaptive super network to generate global image quality discrimination parameters;

[0073] Step 3: Input the local features of the multiple levels into a target prediction network constructed based on the global image quality discrimination parameter to obtain an image quality score of the power image to be evaluated.

[0074] Specifically, before step 1, a network for extracting image semantic features is pre-built. The purpose of image semantic feature extraction is to obtain mathematical expressions of local features and global features in an image. Usually, a complete image can be regarded as being composed of many fragments. The basic feature in each fragment is the line. The line features of the fragments in the image are local features of the image, also called low-level local features. A more structured and complex series of lines in the entire image are high-level features of the image. The high-level features are further abstracted from the low-level features to represent the global features of the image. In the present invention application, a stack of convolutional layers and pooling layers of a neural network is used to extract local and global features. The deeper the network stack, the higher the level of the extracted image features, and the better the effect of image semantic expression.

[0075] When building an image semantic feature extraction network, first build the most basic convolution stacking module. Figure 2As shown, the convolution stacking module consists of a series of convolutional layers for learning local features of images. The structure of the convolution stacking module is: the first 1x1 convolutional layer (1*1Convolve), the 3x3 convolutional layer (3*3Convolve), the second 1x1 convolutional layer (1*1Convolve) and the batch normalization layer (Batch Norm layer) after each convolutional layer are connected in sequence. The first 1x1 convolutional layer is the input end of the convolution stacking module, which is used to reduce the number of channels and reduce the computational complexity; the second 1x1 convolutional layer is the output end of the convolution stacking module, which is used to restore the number of channels and keep the dimension of the feature map consistent. The 3x3 convolutional layer is used to learn local features. The batch normalization layer (Batch Normalization layer) added after each convolutional layer converts the feature map of a batch of data into a distribution that satisfies mean = 0 and variance = 1 through regularization, which helps to alleviate the problems of gradient disappearance and gradient explosion and improve the stability and convergence speed of the network.

[0076] In the present application, the image semantic feature extraction network stacks a preset number of convolutional stacking modules and pooling layers in sequence to extract local features and global features at multiple levels. The number of convolutional stacking modules is preset according to the hardware conditions and the requirements of image evaluation accuracy. The more convolutional stacking modules there are, the richer the feature extraction is. As the number of stacking layers increases, each layer of convolutional stacking modules can extract higher-level local features in the image. However, the increase in the number of stacking layers also increases the hardware requirements and computing time. In the disclosed embodiment, four layers of convolutional stacking modules are connected to form a deep network structure, namely the image semantic feature extraction network in the present invention. The specific structure is as follows: Figure 3 As shown, it is used to learn local and global features of multiple levels of images.

[0077] The process of extracting local features and global features of the power image to be evaluated by the image semantic feature extraction network is as follows: the input of the first convolutional stacking module is used as the input of the image semantic feature extraction network to extract and output the first layer of local features of the power image to be evaluated; the output of the first convolutional stacking module is used as the input of the first pooling layer, and after normalizing the first layer of local features, it is input to the second convolutional stacking module to extract and output the second layer of local features; after the local features of the second layer pass through the next pooling layer, they are input to the next convolutional stacking module to extract and output the local features of the next layer, until the last convolutional stacking module extracts the local features of the highest level as the global features, and the local features and global features of multiple levels output by the convolutional stacking modules are used as the output of the image semantic feature extraction network.

[0078] The main function of the pooling layer between each convolutional stacking module is to provide feature invariance, including image translation invariance, image rotation invariance, and image scaling invariance. The addition of the pooling layer can allow image features to have a certain offset in space, thereby enhancing the robustness of the features.

[0079] At present, in the analysis of power images, the problem of image quality evaluation methods based on traditional machine learning is that it is necessary to construct indicators related to image quality, and the construction of these indicators is relatively professional, and the selection of indicators is different for different scenes and different image quality problems. For example, for image blur, it is necessary to consider the gradient of the line neighborhood in the image, and for the image noise problem, it is necessary to consider whether the distribution of image pixels changes. Therefore, the rationality of feature construction will directly affect the effect of image quality evaluation, and it is necessary to rely on manual collection and annotation of images. In the process of sample data collection, it is usually found that there are a large number of sample images with blur, noise, tilt and other sample data quality problems that are not high or even cannot be recognized manually, which greatly hinders the efficiency of image sample collection and quality inspection. The present invention designs an image semantic feature extraction network, which deepens the level of image feature extraction by repeatedly applying convolution stacking modules, realizes the extraction of multi-level features, wherein the lower network layer extracts low-level features, the higher network layer extracts high-level features, and the highest network extracts the global features of the image, and by adding pooling layers between different levels, the stability of features for image translation transformation is increased. In the subsequent image evaluation, the image is evaluated based on the extracted multiple features, without the need to construct corresponding evaluation indicators based on the image category. This solves the problem of difficult indicator construction and improves the generalization and practicality of the model.

[0080] Specifically, in step 1, the acquired power image to be evaluated is input into the image semantic feature extraction network built based on the convolution stacking module. Through multiple stacked convolution stacking modules and pooling layers, local features and global features at multiple levels can be well extracted. The local features at multiple levels are equivalent to the role of evaluation indicators in traditional evaluation methods. The local features at the highest level are used as global features and are used to generate global image quality discrimination parameters in subsequent solutions. In the future, the local features can be evaluated more closely to the features of the image itself, thereby improving the accuracy of the model.

[0081] Specifically, before step 2, an adaptive super network needs to be constructed first.

[0082] The traditional machine learning method can be expressed as follows:

[0083] y1=f1(x,θ)

[0084] Among them: y1 is the prediction target of the common image, here it refers to the image quality score; f1 is the model function; x is the input image; θ represents the model parameters.

[0085] After training the model with sample data, the model learns the patterns in the sample data and obtains the model parameters θ. When making predictions, for each input image x, the same model parameters θ are used to calculate and obtain the normal image prediction value y1.

[0086] In the power industry, due to the complexity of the business, the background, entities, people, behaviors, etc. involved in the image are very rich and very different. Therefore, the image quality is related to both the local distortion of the image and the overall content. Using the same model parameter θ to judge the quality of images with completely different contents cannot cover different image structures. In order to imitate the human perception process and make full use of the semantic information of the image, the image quality prediction model is expressed as follows:

[0087] y2=f2(x,θ x )

[0088] Where: y2 is the power image prediction target; f2 is the model function; x is the input image; θ x is the power image model parameter. Parameter θ in the model x Relying on each image itself, rather than using the same parameters for all input power images. Since the power image content is different, the perceived image quality is also different. In this way, different features are adaptively obtained for different power image contents, and different parameters are obtained to judge the power image quality.

[0089] In order to train different parameters from different images, the present invention uses an adaptive hyper network to calculate the power image model parameters θ x .

[0090]

[0091] Where: θ x is the power image model parameter; H represents the mapping function of the adaptive hyper network, which is mainly used to learn how to evaluate the quality of different image contents. represents the parameters of the adaptive hyper network, and G(x) represents the global features extracted in step 1. The image quality evaluation model based on the adaptive hyper network follows a top-down perception mechanism. It first uses the global semantic features of the image to generate an overall understanding of the image, and then uses different parameters to evaluate its quality according to different image contents. In this way, the features and judgment rules used for images with different contents will be more flexible.

[0092] Specifically, the present invention constructs an adaptive super network based on the total number of levels of local features extracted in step 1 and the local feature size information of each level. In the embodiment disclosed in the present invention, there are 3 levels of local features that need to be further learned in the future and 1 evaluation based on global information. Therefore, 4 sets of parameters need to be generated as network parameters in the subsequent target prediction network. Therefore, four linear layers are constructed in the adaptive super network. Specifically, the corresponding adaptive super network structure is as follows: Figure 4 As shown in Figure 1, the global features are passed through an adaptive super network with four linear layers to generate four sets of parameters for the subsequent construction of the target prediction network.

[0093] Specifically, in step 2, the global image quality discrimination parameter is generated by using an adaptive super network, including: inputting the global features into an adaptive super network constructed based on local feature related information to generate the global image quality discrimination parameter. In the present invention, the global image quality discrimination parameter will be used as a network parameter to construct a target prediction network. The generation of the global image quality discrimination parameter can be understood as the establishment of a perceptual rule based on the global information of the image.

[0094] In the present invention, the global image quality discrimination parameters include: the local feature global evaluation parameters of each level and the global feature evaluation parameters. The local feature global evaluation parameters of each level are used as the network parameters of the convolution layer of the corresponding level in the subsequent target prediction network. The global feature evaluation parameters are used as the parameters for calculating the final overall image quality score in the subsequent target prediction network. The global image quality discrimination parameters include: the local feature global evaluation parameters of each level and the global feature evaluation parameters.

[0095] At present, for any input image, the model trained by the machine learning algorithm uses the same parameters for quality evaluation. However, for images with different contents, the quality judgment rules are different. For example, for a landscape image with the sky as the background, all the scenery should be clear, while for a macro-photographed insect image, the background is often blurred. Therefore, the same judgment rule is used for image quality evaluation, which cannot adapt to scenes with relatively complex business content. In the present invention, a generation mechanism for the target prediction network parameters of power image quality evaluation is designed. The global features of the power image to be evaluated extracted in step 1 are used to generate different quality discrimination rules for power images with different contents by using an adaptive super network, and the rules are applied to subsequent quality evaluation. The quality discrimination rules are global image quality evaluation parameters. The image quality evaluation parameter generation mechanism based on the adaptive super network architecture proposed in the present invention solves the problem that the evaluation rules are solidified and cannot adapt to relatively complex business scenarios in the prior art, so that the image evaluation model has better applicability and generalization, and meets the needs of high-quality image screening in complex power business scenarios.

[0096] Specifically, in step 3, the target prediction network is used to achieve the final prediction task. The input of the network is the local features of multiple levels of the power image to be evaluated extracted in step 1. Since image quality evaluation mainly analyzes the degree of image distortion, and distortion is mainly reflected in local features, the input of the image quality assessment target prediction network is the local features of multiple levels generated by the above feature extraction network, and the output of the network is the image quality score.

[0097] In this embodiment, taking the extracted local features as 3 levels as an example, the structure of the target prediction network is as follows: Figure 5 As shown, it includes 3 levels of convolutional layers (7*7Convolve), 3 activation function layers (Sigmoid), a connection layer (Concat) and a linear mapping layer (Linear). In the present invention, the number of convolutional layers in the target network depends on the number of local features extracted by the image semantic feature extraction network, and there is a one-to-one correspondence between the convolutional layers and the local features.

[0098] The process of evaluating the power image to be evaluated by the target prediction network is as follows: the input of the convolutional layers of the multiple levels is used as the input of the target prediction network; the convolutional layer of each level is used to further learn and compress the local features of the corresponding level of the input based on the global feature information, obtain the local features of the multiple levels after convolution, and output them to the corresponding activation function layer. In the embodiment of the present disclosure, the convolutional layer is a 7x7 convolutional layer, and the mathematical expression of the local features of each level after convolution is:

[0099]

[0100] Where: L′ i represents the local features of the i-th layer after convolution; L i is the local feature of the i-th layer, W i is the i-th group of parameters. In the specific embodiment of the present invention, a 7x7 convolutional layer is used, so the values ​​of n and m are 1 to 7. Three levels of local features are extracted, so the value of i is 1 to 3.

[0101] The activation function layer in the present invention is used to map the local features after convolution, and it uses a nonlinear activation function to map the value to a range between 0 and 1. A value close to 1 indicates a high activation, and a value close to 0 indicates an inhibition state.

[0102]

[0103] Among them: sigmoid(x1) represents the nonlinear activation function; x1 is the local feature of each level after convolution, that is, the output of the convolution layer. The activation function layer maps the result to between 0 and 1 and then outputs it to the connection layer.

[0104] The connection layer is used to fuse the output results of multiple activation function layers, and then input the fused local features into the linear mapping layer for calculation.

[0105] The linear mapping layer calculates the image quality score through a weighted algorithm based on the global image quality discrimination parameter as the output of the target prediction network. Among them, the global image quality discrimination parameter includes: the local feature global evaluation parameter of each level and the global feature evaluation parameter. The expression of the image quality score is as follows:

[0106] Y=L concat W n +b n

[0107] Where: Y is the image quality score; W n is the weight parameter for global feature evaluation; b n Bias parameter for global feature evaluation; L concat is the fused local feature.

[0108] The weight parameters in the convolution layer and the linear mapping layer in the target prediction network are learned from the image by an adaptive hypernetwork. Specifically, the weight parameters of each convolution layer are determined based on the global evaluation parameters of the local features of each level, and the weight parameters of the linear mapping layer are determined based on the global feature evaluation parameters.

[0109] The overall image quality assessment model can be described as:

[0110]

[0111] Where: Y is the image quality score; L(x) represents the local semantic features extracted from the power image to be evaluated, Represents the global image quality discrimination parameter generated by the adaptive hypernetwork, where the global evaluation parameters of the local features at each level convolve L(x), and the global feature evaluation parameters linearly map the fused features.

[0112] The power image quality assessment method based on the adaptive super network architecture provided by the present invention is tested using images of business scenarios such as power safety supervision and power transmission. The results show that the average accuracy of the model in image quality assessment reaches 80%, which meets the needs of power business image screening and image sample set construction. Compared with the previous image quality assessment based on traditional machine learning, the evaluation effect of this method on power business images has been improved by 20%, which has obvious advantages over traditional models.

[0113] The overall image quality evaluation model construction structure disclosed in the present invention is as follows Figure 6 As shown, it mainly includes three parts: image semantic feature extraction network, adaptive super network and target prediction network, and the corresponding functions are semantic feature extraction, perception rule establishment and image quality prediction respectively. The power image quality assessment method based on the adaptive super network architecture disclosed in the present invention adopts a deep neural network model to extract multi-level local features and global features of power images, and uses an adaptive super network architecture model to generate perception rules for image quality evaluation from global features, which are used to judge the local features of the image based on global features to obtain image quality evaluation scores. This method applies intelligent image processing and discrimination technology, deeply explores the relationship between image content and quality evaluation, and provides a more practical and efficient technical means for the screening of high-quality power images and the construction of image sample sets, promoting the application of power image resources and the construction of image sample sets.

[0114] Embodiment 2:

[0115] Based on the same inventive concept, the present invention also proposes a power image quality assessment system based on an adaptive super network architecture, the system structure is as follows Figure 7 As shown, including:

[0116] Feature extraction module: used to input the acquired power image to be evaluated into a pre-built image semantic feature extraction network to extract local features and global features at multiple levels;

[0117] Parameter generation module: used to generate global image quality discrimination parameters based on the global features using an adaptive super network;

[0118] Image quality assessment module: used to input the local features of the multiple levels into the target prediction network constructed based on the global image quality discrimination parameters to obtain the image quality score of the power image to be evaluated.

[0119] Preferably, the feature extraction module is specifically used for:

[0120] Input the acquired power image to be evaluated into a pre-built image semantic feature extraction network;

[0121] Through multiple convolutional stacking modules connected in series in the image semantic feature extraction network, local features of multiple levels are extracted in sequence, and the local features of the highest level extracted by the last convolutional stacking module are used as global features.

[0122] Preferably, the global image quality discrimination parameters include: local feature global evaluation parameters of each level and global feature evaluation parameters;

[0123] The parameter generation module is specifically used for:

[0124] Building an adaptive super network based on the total number of levels of the local features and the local feature size information of each level;

[0125] The global features are input into the adaptive super network to generate the local feature global evaluation parameters of each level and the global feature evaluation parameters.

[0126] Preferably, the target prediction network in the image quality assessment module includes multiple levels of convolutional layers, multiple activation function layers, connection layers and linear mapping layers;

[0127] Using the inputs of the convolutional layers of the multiple levels as inputs of the target prediction network;

[0128] The convolution layer of each level is used to further learn and compress the local features of the corresponding level of the input based on the global feature information, obtain the local features of multiple levels after convolution, and output them to the corresponding activation function layer;

[0129] The activation function layer is used to map the local features after convolution and output them to the connection layer;

[0130] The connection layer is used to fuse the output results of multiple activation function layers, and then input the fused local features into the linear mapping layer for calculation to obtain an image quality score as the output of the target prediction network;

[0131] The weight parameters of each convolutional layer are determined based on the global evaluation parameters of the local features of each level, and the weight parameters of the linear mapping layer are determined based on the global feature evaluation parameters.

[0132] Preferably, the global feature evaluation parameters in the image quality assessment module include: a global feature evaluation weight parameter and a global feature evaluation bias parameter; the expression of the image quality score is as follows:

[0133] Y=L concat W n +b n

[0134] Where: Y is the image quality score; W n is the weight parameter for global feature evaluation; b n Bias parameter for global feature evaluation; L concat is the fused local feature.

[0135] Preferably, the image semantic feature extraction network in the feature extraction module includes: a preset number of convolution stacking modules and pooling layers, wherein the convolution stacking modules and corresponding pooling layers are stacked in sequence;

[0136] The input of the first convolution stack module is used as the input of the image semantic feature extraction network to extract and output the first layer of local features of the power image to be evaluated;

[0137] The output of the first convolution stack module is used as the input of the first pooling layer. After normalizing the local features of the first layer, it is input into the second convolution stack module to extract and output the local features of the second layer.

[0138] After the local features of the second layer pass through the next pooling layer, they are input into the next convolutional stacking module to extract and output the local features of the next layer, until the last convolutional stacking module extracts the local features of the highest level as global features, and then the local features and global features of multiple levels output by each convolutional stacking module are used as the output of the image semantic feature extraction network.

[0139] Preferably, the convolution stacking module in the feature extraction module includes a first 1x1 convolution layer, a second 1x1 convolution layer, a 3x3 convolution layer and a batch normalization layer connected in sequence, wherein the first 1x1 convolution layer is the input end of the convolution stacking module, the second 1x1 convolution layer is the output end of the convolution stacking module, the 3x3 convolution layer is used to extract local features, and the batch normalization layer is used to regularize the output of each convolution layer.

[0140] The present invention realizes the extraction of local features and global features at multiple levels through the feature extraction module, parameter generation module and image quality assessment module in the power image quality assessment system based on the adaptive super network architecture, generates network parameters in the target prediction network based on the global features, and then evaluates the image based on the local features through the target prediction network, thereby improving the generalization of the model and the accuracy of the assessment.

[0141] Example 3

[0142] like Figure 8 As shown, the present invention also provides an electronic device, which may be a computer device, a single-chip device, an intelligent mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, the processor, and the transceiver component are connected via a bus; the memory may be used to store an execution program, and an exemplary execution program may include instructions; the processor is used to execute the instructions stored in the memory. The memory may also be used to store data, which may be called and / or modified when the instructions are executed.

[0143] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in a storage medium to implement the corresponding method flow or corresponding functions, so as to implement the steps of a power image quality assessment method based on an adaptive super-network architecture in the above-mentioned embodiment.

[0144] Example 4

[0145] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in an electronic device for storing programs and data. It can be understood that the storage medium here can include both built-in storage media in electronic devices and extended storage media supported by electronic devices. The storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by a processor are also stored in the storage space, and these instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor loads and executes one or more instructions stored in the storage medium, which can implement the steps of a power image quality assessment method based on an adaptive super network architecture in the above embodiment.

[0146] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0147] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0148] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1A function specified in one or more boxes.

[0149] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit its protection scope. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that after reading the present invention, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the protection scope of the claims to be approved.

Claims

1. A method for evaluating power image quality based on an adaptive super network architecture, characterized in that: include: The acquired power image to be evaluated is input into a pre-built image semantic feature extraction network to extract local features and global features at multiple levels; Based on the global features, using an adaptive super network to generate global image quality discrimination parameters; The local features of the multiple levels are input into a target prediction network constructed based on the global image quality discrimination parameter to obtain an image quality score of the power image to be evaluated.

2. The method according to claim 1, characterized in that The acquired power image to be evaluated is input into a pre-built image semantic feature extraction network to extract local features and global features at multiple levels, including: Input the acquired power image to be evaluated into a pre-built image semantic feature extraction network; Through multiple convolutional stacking modules connected in series in the image semantic feature extraction network, local features of multiple levels are extracted in sequence, and the local features of the highest level extracted by the last convolutional stacking module are used as global features.

3. The method according to claim 2, characterized in that The global image quality discrimination parameters include: local feature global evaluation parameters of each level and global feature evaluation parameters; The method of generating global image quality discrimination parameters based on the global features by using an adaptive super network includes: Building an adaptive super network based on the total number of levels of the local features and the local feature size information of each level; The global features are input into the adaptive super network to generate the local feature global evaluation parameters of each level and the global feature evaluation parameters.

4. The method according to claim 3, characterized in that The target prediction network includes multiple levels of convolutional layers, multiple activation function layers, connection layers and linear mapping layers; Using the inputs of the convolutional layers of the multiple levels as inputs of the target prediction network; The convolution layer of each level is used to further learn and compress the local features of the corresponding level of the input based on the global feature information, obtain the local features of multiple levels after convolution, and output them to the corresponding activation function layer; The activation function layer is used to map the local features after convolution and output them to the connection layer; The connection layer is used to fuse the output results of multiple activation function layers, and then input the fused local features into the linear mapping layer for calculation to obtain an image quality score as the output of the target prediction network; The weight parameters of each convolutional layer are determined based on the global evaluation parameters of the local features of each level, and the weight parameters of the linear mapping layer are determined based on the global feature evaluation parameters.

5. The method according to claim 4, characterized in that The global feature evaluation parameters include: a global feature evaluation weight parameter and a global feature evaluation bias parameter; the expression of the image quality score is as follows: Y=L concat W n +b n Where: Y is the image quality score; W n is the weight parameter for global feature evaluation; b n Bias parameter for global feature evaluation; L concat is the fused local feature.

6. The method according to claim 1, characterized in that The image semantic feature extraction network includes: a preset number of convolution stacking modules and pooling layers, wherein the convolution stacking modules and corresponding pooling layers are stacked in sequence; The input of the first convolution stack module is used as the input of the image semantic feature extraction network to extract and output the first layer of local features of the power image to be evaluated; The output of the first convolution stack module is used as the input of the first pooling layer. After normalizing the local features of the first layer, it is input into the second convolution stack module to extract and output the local features of the second layer. After the local features of the second layer pass through the next pooling layer, they are input into the next convolutional stacking module to extract and output the local features of the next layer, until the last convolutional stacking module extracts the local features of the highest level as global features, and then the local features and global features of multiple levels output by each convolutional stacking module are used as the output of the image semantic feature extraction network.

7. The method according to claim 6, characterized in that The convolution stacking module includes a first 1x1 convolution layer, a second 1x1 convolution layer, a 3x3 convolution layer, and a batch normalization layer connected in sequence, wherein the first 1x1 convolution layer is the input end of the convolution stacking module, the second 1x1 convolution layer is the output end of the convolution stacking module, the 3x3 convolution layer is used to extract local features, and the batch normalization layer is used to regularize the output of each convolution layer.

8. A power image quality assessment system based on an adaptive super network architecture, characterized in that: include: Feature extraction module: used to input the acquired power image to be evaluated into a pre-built image semantic feature extraction network to extract local features and global features at multiple levels; Parameter generation module: used to generate global image quality discrimination parameters based on the global features using an adaptive super network; Image quality assessment module: used to input the local features of the multiple levels into the target prediction network constructed based on the global image quality discrimination parameters to obtain the image quality score of the power image to be evaluated.

9. The system according to claim 8, characterized in that The feature extraction module is specifically used for: Input the acquired power image to be evaluated into a pre-built image semantic feature extraction network; Through multiple convolutional stacking modules connected in series in the image semantic feature extraction network, local features of multiple levels are extracted in sequence, and the local features of the highest level extracted by the last convolutional stacking module are used as global features.

10. The system according to claim 9, characterized in that The global image quality discrimination parameters include: local feature global evaluation parameters of each level and global feature evaluation parameters; The parameter generation module is specifically used for: Building an adaptive super network based on the total number of levels of the local features and the local feature size information of each level; The global features are input into the adaptive super network to generate the local feature global evaluation parameters of each level and the global feature evaluation parameters.

11. The system according to claim 10, characterized in that The target prediction network in the image quality assessment module includes multiple levels of convolutional layers, multiple activation function layers, connection layers and linear mapping layers; Using the inputs of the convolutional layers of the multiple levels as inputs of the target prediction network; The convolution layer of each level is used to further learn and compress the local features of the corresponding level of the input based on the global feature information, obtain the local features of multiple levels after convolution, and output them to the corresponding activation function layer; The activation function layer is used to map the local features after convolution and output them to the connection layer; The connection layer is used to fuse the output results of multiple activation function layers, and then input the fused local features into the linear mapping layer for calculation to obtain an image quality score as the output of the target prediction network; The weight parameters of each convolutional layer are determined based on the global evaluation parameters of the local features of each level, and the weight parameters of the linear mapping layer are determined based on the global feature evaluation parameters.

12. The system according to claim 8, characterized in that The global feature evaluation parameters in the image quality assessment module include: a global feature evaluation weight parameter and a global feature evaluation bias parameter; the expression of the image quality score is as follows: Y=L concat W n +b n Where: Y is the image quality score; W n is the weight parameter for global feature evaluation; b n Bias parameter for global feature evaluation; L concat is the fused local feature.

13. The system according to claim 8, characterized in that The image semantic feature extraction network in the feature extraction module includes: a preset number of convolution stacking modules and pooling layers, wherein the convolution stacking modules and corresponding pooling layers are stacked in sequence; The input of the first convolution stack module is used as the input of the image semantic feature extraction network to extract and output the first layer of local features of the power image to be evaluated; The output of the first convolution stack module is used as the input of the first pooling layer. After normalizing the local features of the first layer, it is input into the second convolution stack module to extract and output the local features of the second layer. After the local features of the second layer pass through the next pooling layer, they are input into the next convolutional stacking module to extract and output the local features of the next layer, until the last convolutional stacking module extracts the local features of the highest level as global features, and then the local features and global features of multiple levels output by each convolutional stacking module are used as the output of the image semantic feature extraction network.

14. The system according to claim 13, characterized in that The convolution stacking module in the feature extraction module includes a first 1x1 convolutional layer, a second 1x1 convolutional layer, a 3x3 convolutional layer, and a batch normalization layer connected in sequence, wherein the first 1x1 convolutional layer is the input end of the convolution stacking module, the second 1x1 convolutional layer is the output end of the convolution stacking module, the 3x3 convolutional layer is used to extract local features, and the batch normalization layer is used to regularize the output of each convolutional layer.

Images