An electric power scene image data automatic screening method

Abstract

The application discloses an electric power scene image data automatic screening method, belongs to the technical field of image screening, and aims to provide an electric power scene image data automatic screening method, solves the problems of great labor intensity, low screening efficiency and low screening accuracy of existing electric power sample data depending on manual screening. The combination of image classification, target detection and image retrieval improves the prediction accuracy of image categories, effectively realizes the feature extraction capability of the data screening network, greatly improves the screening speed and screening accuracy, and ensures the screening efficiency of the electric power scene image data. The application is suitable for the electric power scene image data automatic screening method.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of image screening, and particularly relates to an electric power scene image data automatic screening method. BACKGROUND

[0002] In recent years, the power grid is an efficient and fast energy transmission channel and an optimized configuration platform, is a key link of sustainable development of energy and power, and plays an important hub role in the modern energy supply system, and is related to national energy security. Hidden dangers have always been the "prologue" of major risk accidents. In recent years, the number of daily on-site operation points and on-site operation personnel of power companies has been increasing. There are many risks in the face of a large number of people, and it is obviously not practical to rely on the traditional "person-to-person" supervision mode to find hidden dangers, so it is necessary to eliminate risks in the "germination" while taking into account the efficiency.

[0003] In the electric power scene, due to the large amount of data and the large amount of data of electric power big data, manual screening is often used to screen and process repeated and non-compliant scene data.

[0004] However, the manual screening method has the following three shortcomings:

[0005] (1) In the case of a large amount of data, the labor intensity of the employees is high, and the screening cost is high;

[0006] (2) The manual screening efficiency is low;

[0007] (3) The screening has personal subjectivity and no specific quantitative evaluation standard.

[0008] Therefore, the automatic screening and visualization of electric power sample data have great significance for reducing the labor intensity of employees, improving the processing efficiency, accelerating the construction of an international leading energy internet enterprise with Chinese characteristics, and showing new achievements in building a new development pattern and serving the party and state work. SUMMARY

[0009] The purpose of the present application is to provide an electric power scene image data automatic screening method, which solves the problems of high labor intensity, low screening efficiency and low screening accuracy of existing electric power sample data relying on manual screening.

[0010] The technical solution adopted by the present application is as follows:

[0011] An electric power scene image data automatic screening method comprises the following steps:

[0012] (1) inputting an image sample to be screened;

[0013] (2) positioning the object area through the Mask R-CNN framework of the target detection algorithm model;

[0014] (3) The Mask R-CNN framework obtains the highest confidence feature map of different regions through non-maximum suppression;

[0015] (4) Detecting the feature map of different regions of the input image sample to be screened, and searching all other pictures existing the feature map in the database;

[0016] (5) The pictures existing the feature map in the database are sorted according to the similarity of the region where the feature map is located in the input image sample to be screened, and the sorted pictures are sorted; and the automatic screening of the image data is completed.

[0017] Further, in step (3), based on the feature map of the image sample to be screened, the image classification of the input image sample to be screened is performed through the picture classification algorithm.

[0018] Further, in the image classification process, the category probability of the image sample to be screened is used to determine whether the picture is compliant data, and if it is non-compliant data, the database is searched to query the category to which the nearest neighbor picture of the image sample to be screened belongs, as the image category of the image sample to be screened.

[0019] Further, in step (5), the step of sorting the retrieved pictures is:

[0020] (a) Based on the consistency of the image category of the image sample to be screened, the first sorting is performed;

[0021] (b) Based on the number of feature maps, the second sorting is performed;

[0022] (c) Based on the similarity of the region where the feature map is located in the image sample to be screened, the third sorting is performed.

[0023] Further, in step (1), the input image sample to be screened is preprocessed.

[0024] Further, the preprocessing of the image sample to be screened includes the following steps:

[0025] S1, optical correction is realized based on the histogram equalization method of non-linear gray scale transformation;

[0026] S2, denoising processing is performed based on the wavelet denoising algorithm.

[0027] As described above, due to the adoption of the above technical solutions, the present application has the following advantages:

[0028] 1. In the present invention, a histogram equalization method based on nonlinear grayscale transformation is used to achieve optical correction of the image samples to be screened, thereby achieving the purpose of image enhancement and contrast improvement; based on the wavelet denoising algorithm, the phenomenon of overexposure and underexposure of the image is solved, ensuring the accuracy of subsequent screening.

[0029] 2. In the present invention, image classification, target detection and image retrieval are combined, which not only improves the prediction accuracy of image categories but also effectively realizes the feature extraction capability of data screening network, greatly improving the screening speed and accuracy, and ensuring the screening efficiency of power scene image data.

[0030] 3. In the present invention, the Mask R-CNN image recognition technology is used to ensure that features can be quickly and accurately identified in complex and changing backgrounds, fully balancing the recognition accuracy and speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be considered as limiting the scope. A person of ordinary skill in the art can also derive other relevant drawings based on these drawings without inventive effort, among which:

[0032] Figure 1 It is a schematic diagram of the overall process of the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0034] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are intended to fall within the scope of protection of the present invention.

[0035] A method for automatically screening power scene image data includes the following steps:

[0036] (1) Input the image sample to be screened;

[0037] (2) Locate the object area through the Mask R-CNN framework of the target detection algorithm model;

[0038] (3) The Mask R-CNN framework obtains the highest confidence feature map of different regions through non-maximum suppression;

[0039] (4) Detecting the feature map of different regions of the input image sample to be screened, and searching all other pictures existing the feature map in the database;

[0040] (5) The pictures existing the feature map in the database are sorted according to the similarity of the region where the feature map is located in the input image sample to be screened, and the sorted pictures are sorted according to the similarity of the region where the feature map is located in the input image sample to be screened.

[0041] Further, in step (3), based on the feature map of the image sample to be screened, the image classification algorithm is used to classify the image category of the input image sample to be screened.

[0042] Further, in the image classification process, the category probability of the image sample to be screened is used to determine whether the picture is compliant data, and if it is not compliant data, the database is searched to query the category of the nearest neighbor picture of the image sample to be screened as the image category of the image sample to be screened.

[0043] Further, in step (5), the step of sorting the retrieved pictures is:

[0044] (a) Based on the consistency of the image category of the image sample to be screened, the sorting is performed once;

[0045] (b) Based on the number of feature maps, the sorting is performed twice;

[0046] (c) Based on the similarity of the region where the feature map is located in the image sample to be screened, the sorting is performed three times.

[0047] Further, in step (1), the input image sample to be screened is preprocessed.

[0048] Further, the preprocessing of the image sample to be screened includes the following steps:

[0049] S1, the histogram equalization method based on nonlinear gray scale transformation is used to realize optical correction;

[0050] S2, the wavelet denoising algorithm is used for denoising processing.

[0051] In the implementation process of the present application:

[0052] First, the input image sample to be screened is preprocessed, and the histogram equalization method based on nonlinear gray scale transformation is used to realize optical correction, and then the wavelet denoising algorithm is used for denoising processing.

[0053] locating the object region through a Mask R-CNN framework of the target detection algorithm model; the Mask R-CNN framework obtains the feature map with the highest confidence in different regions through non-maximum suppression; and based on the feature map of the image sample to be screened, the image sample to be screened is classified into an image category through an image classification algorithm.

[0054] In the image classification process, whether the image is compliant data is determined based on the category probability of the image sample to be screened, if the image is non-compliant data, the database is searched to query the category to which the nearest neighbor image of the image sample to be screened belongs, as the image category of the image sample to be screened.

[0055] Then, the feature map of different regions of the input image sample to be screened is detected, and all other images existing the feature map in the database are searched; the images existing the feature map in the database are sorted according to the similarity of the region where the feature map in the input image sample to be screened is located.

[0056] The step of sorting the searched images is:

[0057] (a) based on the consistency of the image category of the image sample to be screened, sorting once;

[0058] (b) based on the number of feature maps, sorting twice;

[0059] (c) based on the similarity of the region where the feature map in the image sample to be screened is located, sorting three times.

[0060] Finally, the automatic screening of image data is completed.

[0061] Embodiment 1

[0062] An automatic screening method of power scene image data, comprising the following steps:

[0063] (1) inputting an image sample to be screened;

[0064] (2) locating the object region through a Mask R-CNN framework of the target detection algorithm model;

[0065] (3) the Mask R-CNN framework obtains the feature map with the highest confidence in different regions through non-maximum suppression;

[0066] (4) detecting the feature map of different regions of the input image sample to be screened, and searching all other images existing the feature map in the database;

[0067] (5) sorting the images existing the feature map in the database according to the similarity of the region where the feature map in the input image sample to be screened is located; and completing the automatic screening of image data.

[0068] Embodiment 2

[0069] On the basis of Embodiment 1, in step (3), based on the feature map of the image sample to be screened, the image category classification of the input image sample to be screened is performed by the picture classification algorithm.

[0070] Embodiment 3

[0071] On the basis of the above-mentioned embodiments, in the image classification process, whether the picture is compliant data is judged based on the category probability of the image sample to be screened, and if it is non-compliant data, the database is searched to query the category to which the nearest neighbor picture of the image sample to be screened belongs, as the image category of the image sample to be screened.

[0072] Embodiment 4

[0073] On the basis of the above-mentioned embodiments, in step (5), the step of sorting the searched pictures is:

[0074] (a) based on the consistency with the image category of the image sample to be screened, the first sorting is performed;

[0075] (b) based on the number of feature maps, the second sorting is performed;

[0076] (c) based on the similarity with the region where the feature map in the image sample to be screened is located, the third sorting is performed.

[0077] Embodiment 5

[0078] On the basis of the above-mentioned embodiments, in step (1), the input image sample to be screened is preprocessed.

[0079] Embodiment 6

[0080] On the basis of the above-mentioned embodiments, the preprocessing of the image sample to be screened includes the following steps:

[0081] S1, the histogram equalization method based on the nonlinear gray scale transformation is used to realize the optical correction;

[0082] S2, the denoising processing is performed based on the wavelet denoising algorithm.

[0083] The above is the embodiment of the present application. The foregoing is the various preferred embodiments of the present application, and the preferred embodiments in the various preferred embodiments can be arbitrarily superimposed and combined if not obviously contradictory or with a certain preferred embodiment as a prerequisite. The embodiments and specific parameters in the embodiments are only for clearly describing the verification process of the application and are not used to limit the patent protection scope of the application. The patent protection scope of the application is still subject to its claims. Any equivalent changes made by using the content of the specification and drawings of the application should also be included in the protection scope of the application.

Claims

1. A method for automatically screening electric power scene image data, characterized by, The method comprises the following steps: (1) inputting an image sample to be screened; (2) locating an object region by a Mask R-CNN framework of a target detection algorithm model; (3) obtaining a feature map with the highest confidence in different regions by the Mask R-CNN framework through non-maximum suppression; (4) detecting the feature map of different regions of the input image sample to be screened, and searching all other pictures with the feature map in a database; (5) sorting the pictures with the feature map in the database according to the similarity of the region where the feature map is located in the input image sample to be screened; completing automatic screening of image data, and the step of sorting the searched pictures comprises: (a) sorting once based on the consistency of the image category of the image sample to be screened; (b) sorting twice based on the number of feature maps; (c) sorting thrice based on the similarity of the region where the feature map is located in the image sample to be screened.

2. The method of claim 1, wherein the method further comprises: In the step (3), the image category of the input image sample to be screened is classified by a picture classification algorithm based on the feature map of the image sample to be screened.

3. The method of claim 2, wherein the method further comprises: In the image classification process, whether the picture is compliant data is determined based on the category probability of the image sample to be screened; if the picture is non-compliant data, the database is searched, and the category to which the nearest neighbor picture of the image sample to be screened belongs is queried as the image category of the image sample to be screened.

4. The method of claim 1, wherein the power scene image data is automatically filtered by: In the step (1), the input image sample to be screened is preprocessed.

5. The method of claim 4, wherein the step of automatically filtering the power scene image data is performed by a computer program. The preprocessing of the image sample to be screened comprises the following steps: S1, realizing optical correction based on a histogram equalization method of a nonlinear gray scale transformation; S2, performing denoising processing based on a wavelet denoising algorithm.

Images