CN117197092A - Underground coal mine image quality assessment method - Google Patents
Underground coal mine image quality assessment method Download PDFInfo
- Publication number
- CN117197092A CN117197092A CN202311175624.2A CN202311175624A CN117197092A CN 117197092 A CN117197092 A CN 117197092A CN 202311175624 A CN202311175624 A CN 202311175624A CN 117197092 A CN117197092 A CN 117197092A
- Authority
- CN
- China
- Prior art keywords
- image
- model
- image quality
- data
- coal mine
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 239000003245 coal Substances 0.000 title claims abstract description 52
- 238000001303 quality assessment method Methods 0.000 title claims abstract description 28
- 238000000034 method Methods 0.000 claims abstract description 65
- 238000011156 evaluation Methods 0.000 claims abstract description 61
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 60
- 238000012549 training Methods 0.000 claims abstract description 43
- 238000005516 engineering process Methods 0.000 claims abstract description 37
- 238000012360 testing method Methods 0.000 claims abstract description 29
- 238000013441 quality evaluation Methods 0.000 claims description 32
- 238000002790 cross-validation Methods 0.000 claims description 13
- 238000001914 filtration Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 11
- 230000003044 adaptive effect Effects 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 10
- 238000013136 deep learning model Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 7
- 230000000007 visual effect Effects 0.000 claims description 6
- 230000000694 effects Effects 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 5
- 238000013519 translation Methods 0.000 claims description 5
- 238000010200 validation analysis Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 230000010354 integration Effects 0.000 claims description 4
- 238000001228 spectrum Methods 0.000 claims description 4
- 239000002131 composite material Substances 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 abstract description 10
- 238000013461 design Methods 0.000 abstract description 2
- 238000003707 image sharpening Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000012854 evaluation process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 229910052500 inorganic mineral Inorganic materials 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000011707 mineral Substances 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 235000012434 pretzels Nutrition 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of coal mine underground image quality assessment methods, in particular to a coal mine underground image quality assessment method which comprises the following steps of data set preparation, model design and training, image quality assessment and model performance assessment. In the invention, the characteristic of the image is extracted through the Convolutional Neural Network (CNN), the information in the image is captured, the connection is established with the Convolutional Neural Network (CNN) through the image enhancement technology, and the Convolutional Neural Network (CNN) is assisted or improved to extract the image characteristic more accurately so as to comprehensively evaluate the image quality. And performing model training and performance evaluation on the trained model by adopting a data set dividing and independent data set testing method. And the cross verification method is used for verifying the model on different data subsets for multiple times, so that the reliability and generalization capability of model evaluation are improved, and the deviation caused by improper selection of the data sets is reduced. The diversity of the data set is increased through the data enhancement technology, the robustness of the model is improved, the model is adapted to different sample changes, and the over-fitting problem is relieved through the super-parameter tuning and model regularization method.
Description
Technical Field
The invention relates to the technical field of coal mine underground image quality assessment methods, in particular to a coal mine underground image quality assessment method.
Background
The underground coal mine image quality evaluation method is an evaluation type method widely used for evaluating the quality of images acquired in a mine environment in terms of definition, contrast and noise level, and due to the fact that the underground coal mine environment is complex, the light condition is limited, and miners may need to rely on the images to judge and decide when performing tasks. Therefore, the method has great significance in accurately evaluating the quality of the underground acquired image, and can help miners and mineral managers to determine whether the image is clear enough to support the work and safety of the miners, and generally comprises definition evaluation, contrast evaluation, noise evaluation, average value evaluation and resolution evaluation.
The definition refers to the edge definition degree of a target object in an image, and the definition evaluation method can evaluate the definition level of the image by calculating the definition value of each region in the image based on the sharpness and edge gradient index of the image. Contrast refers to the degree of brightness difference in an image, and for a coal mine underground image, the degree of contrast can influence the visibility of a target object in the image, and a contrast evaluation method can determine the contrast level of the image by calculating pixel value differences of different areas in the image. There are often various noise sources in the downhole environment, such as electromagnetic interference, insufficient illumination, which can affect image quality, and noise assessment methods can assess the noise condition of the image by analyzing the noise level in the image, such as gaussian noise, pretzel noise. The color value of each pixel point in the image reflects the quality of the image to a certain extent, the average value evaluation method can evaluate the quality level of the image by calculating the average value of each color channel in the image, the resolution refers to the minimum detail size which can be distinguished in the image, in the underground coal mine environment, the resolution is critical to the visual identification task of miners, and the resolution evaluation method can evaluate the resolution level of the image by calculating the edge or detail definition of an object in the image.
In the actual use process of the existing coal mine underground image quality evaluation method, the existing feature extraction method is possibly limited by the feature representation of manual design, so that complex features of an image are difficult to comprehensively capture, the performance of a model is reduced when a complex image scene, illumination change or noise factors are processed, and the evaluation index is lack of comprehensive measurement, so that the accuracy of a judgment result of image definition, contrast and color assurance is insufficient. Whereas in the case of an overall dataset, the collection of datasets adopts too randomness, resulting in overall performance imbalance, and poor performance in the case of partial subsets, unknown test data.
Disclosure of Invention
The invention aims to solve the defects in the prior art, and provides a coal mine underground image quality evaluation method.
In order to achieve the above purpose, the present invention adopts the following technical scheme: the underground coal mine image quality evaluation method comprises the following steps:
data acquisition, integrating and generating a data set;
designing a convolutional neural network model, and performing model training on the convolutional neural network model based on the data set;
performing an image quality assessment based on the convolutional neural network model;
and performing performance evaluation on the convolutional neural network model.
As a further aspect of the invention, the data set contains coal mine downhole images of various quality levels, and the coal mine downhole images of various quality levels contain common image quality problems including blurring, noise and insufficient contrast;
the step of data acquisition and integration generation of the data set comprises the following steps:
collecting and labeling quality grades and problem labels for large-scale coal mine underground image data sets;
the data enhancement technology is applied, including rotation, scaling and translation, so that the diversity of a data set is increased and the robustness of a model is improved;
the data is enhanced by the integrated data enhancement technology, and the data sets are integrated and classified.
As a further scheme of the present invention, the step of designing a convolutional neural network model and performing model training on the convolutional neural network model based on the data set specifically includes:
designing a framework of a convolutional neural network, and defining a network structure, a convolutional layer, a pooling layer and a full-connection layer to extract image features;
selecting a proper loss function and an optimizer for the convolutional neural network, and performing network training by utilizing the data set after image enhancement processing;
and (5) performing model tuning.
As a further scheme of the invention, the loss function adopts the difference between the absolute error measurement predicted value and the real quality value, the optimization algorithm adopts an Adam optimizer, and the effect of the training process is adjusted based on the mode of adjusting the learning rate parameter.
As a further scheme of the invention, the model tuning specifically means that the optimal hyper-parameter configuration is found by adopting a grid search mode, the complexity of the model is controlled by a regularization method comprising L1 regularization and L2 regularization, the fitting problem is relieved, the performance of the model on different data subsets is evaluated by using cross verification, and the optimal model configuration is selected.
As a further aspect of the present invention, the step of performing image quality assessment based on the convolutional neural network model specifically includes:
preprocessing the data of the image quality assessment by adopting an image enhancement technology;
loading an image quality evaluation algorithm and a feature extraction method based on computer vision and image processing technology, and automatically detecting and identifying image quality problems by analyzing pixel values, spectrum features and texture features of an image;
performing quality evaluation on the underground coal mine image by adopting a deep learning model;
with reference to the multiple sets of image quality assessment indicators, a composite image quality score is calculated.
As a further scheme of the invention, the image enhancement technology comprises noise removal, contrast enhancement and sharpening, wherein the noise removal specifically comprises the adoption of a filtering technology, the noise in the image is reduced based on a median filtering and Gaussian filtering method, the contrast enhancement specifically comprises the adoption of a histogram equalization and contrast stretching method, the contrast and visual effect of the image are improved, and the sharpening specifically comprises the adoption of a Laplace operator to improve the definition of the image.
As a further aspect of the present invention, the image quality evaluation index includes a structural similarity index, a peak signal-to-noise ratio, and an adaptive contrast enhancement;
the combined image quality scores are calculated by a weighted average method, and weights are distributed based on 40% structural similarity index, 20% peak signal to noise ratio and 40% adaptive contrast enhancement.
As a further aspect of the present invention, the step of performing performance evaluation on the convolutional neural network model specifically includes:
adopting a random number generator, taking 80% of data as a training set, 20% of data as a testing set proportion, and dividing the data set into the training set and the testing set;
performing performance evaluation on the trained model by using an independent test data set, wherein evaluation indexes of the performance evaluation comprise accuracy, recall, precision and F1 score;
and performing cross-validation.
As a further scheme of the invention, the specific K-fold cross validation is performed by dividing the data set into K folds, using K-1 folds as training sets each time, using the rest 1 folds as validation sets, repeating K times, and taking the average performance as the comprehensive evaluation of the model.
Compared with the prior art, the invention has the advantages and positive effects that:
according to the invention, the characteristics of the image are extracted through the Convolutional Neural Network (CNN), the information in the image is captured, and the quality of the model to the input image is improved through an image enhancement technology, so that a connection is established with the Convolutional Neural Network (CNN), the Convolutional Neural Network (CNN) is assisted or improved to extract the image characteristics more accurately, and the image quality is comprehensively evaluated by adopting a method of comprehensively considering a plurality of evaluation indexes. And performing model training and performance evaluation on the trained model by adopting a data set dividing and independent data set testing method. The cross verification method is used for verifying the model on different data subsets for multiple times, so that the reliability and generalization capability of model evaluation are further improved, and deviation caused by improper selection of the data sets is reduced. The diversity of the data set is increased through the data enhancement technology, the robustness of the model is improved, the model is adapted to different sample changes, and the model is further optimized through the super-parameter tuning and model regularization method, so that the over-fitting problem is relieved.
Drawings
FIG. 1 is a schematic diagram of main steps of a coal mine underground image quality evaluation method provided by the invention;
FIG. 2 is a detailed flow chart of a step 1 of a coal mine underground image quality evaluation method provided by the invention;
FIG. 3 is a step 2 refinement flow chart of a coal mine underground image quality evaluation method provided by the invention;
FIG. 4 is a step 3 refinement flowchart of a coal mine underground image quality evaluation method provided by the invention;
fig. 5 is a step 4 refined flowchart of the coal mine underground image quality evaluation method provided by the invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
In the description of the present invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like indicate orientations or positional relationships based on the orientation or positional relationships shown in the drawings, merely to facilitate describing the present invention and simplify the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention. Furthermore, in the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.
Example 1
Referring to fig. 1, the present invention provides a technical solution: the underground coal mine image quality evaluation method comprises the following steps:
data acquisition, integrating and generating a data set;
designing a convolutional neural network model, and performing model training on the convolutional neural network model based on a data set;
performing image quality assessment based on the convolutional neural network model;
and performing performance evaluation on the convolutional neural network model.
Specifically, the strong feature extraction capability of the Convolutional Neural Network (CNN) effectively extracts the features of the image by designing a proper network structure, a convolutional layer and a pooling layer, so that the information of the image is better captured, and meanwhile, the image enhancement technology further optimizes the input image, including denoising and contrast increasing operations, improves the input quality of a model, and is beneficial to accurately extracting the features of the image and evaluating the quality of the image.
According to the technical scheme, a plurality of evaluation indexes such as structural similarity indexes, peak signal-to-noise ratio and self-adaptive contrast enhancement are comprehensively considered, the image quality such as definition, contrast and detail reservation can be comprehensively evaluated from multiple aspects through the comprehensive evaluation indexes, and the comprehensive evaluation method can more comprehensively reflect the image quality and provide more accurate evaluation results.
In the aspect of model performance evaluation, the technical scheme adopts data set division and independent test data sets, ensures that the performance evaluation of the model is independent, can truly reflect the generalization capability of the model on unknown data, and further adopts a cross verification method to verify the model on different data subsets, so that the reliability and generalization capability of the evaluation are further improved.
The data enhancement technology increases the diversity of a data set, improves the robustness of a model through rotation, scaling and translation operations, so that the model can adapt to various sample changes, and simultaneously optimizes the complexity of the model and reduces the over-fitting problem through a super-parameter tuning and model regularization method, so that the model is more robust and has generalization capability.
By combining the above, the technical scheme can fully utilize the feature extraction capability of the Convolutional Neural Network (CNN), comprehensively consider a plurality of evaluation indexes, independently and reliably evaluate the performance of the model, has robustness and adjustability, can provide accurate and comprehensive underground coal mine image quality evaluation, can be widely applied in different scenes and data sets, and is expected to further improve the accuracy and practicability of the image quality evaluation through continuous improvement and optimization.
Referring to fig. 2, the data set contains coal mine downhole images of various quality levels, which contain common image quality problems including blurring, noise, and insufficient contrast;
the step of data acquisition and integration to generate a data set comprises the following steps:
collecting and labeling quality grades and problem labels for large-scale coal mine underground image data sets;
the data enhancement technology is applied, including rotation, scaling and translation, so that the diversity of a data set is increased and the robustness of a model is improved;
the data is enhanced by the integrated data enhancement technology, and the data sets are integrated and classified.
Specifically, quality grades and problem labels are marked on a large-scale coal mine underground image dataset image, representative coal mine underground image data is collected through field investigation and support of partners, and the dataset is ensured to contain images with various quality grades, including problems of blurring, noise and insufficient contrast. The image is marked, including quality grade and specific image quality problem, based on computer vision and image processing technology, loading image quality evaluation algorithm and feature extraction method, applying computer vision and image processing technology such as Convolutional Neural Network (CNN), loading image quality evaluation algorithm and feature extraction method, and analyzing pixel value, frequency spectrum feature and texture feature of the image. The image quality problems are automatically detected and identified, the prepared image data sets are marked, the quality grade of each image is marked, such as high quality, general and low quality grade, meanwhile, specific image quality problems are marked, such as blurring, noise and insufficient contrast, and professional staff marks, so that accuracy and consistency are ensured.
Further, the image quality assessment algorithm may be further improved by using a Convolutional Neural Network (CNN), using a pre-trained Convolutional Neural Network (CNN), such as VGG, resNet or EfficientNet, in conjunction with the dataset data for fine tuning to achieve a more accurate quality assessment result;
in the image quality problem label direction, expanding image distortion, color shift and artifact labels, and improving the richness of a data set and the fine granularity of quality assessment;
the image quality evaluation algorithm is combined with the image processing technology to construct an end-to-end system, which can receive the original image as input, automatically detect and identify the image quality problem and give out the corresponding quality grade and problem label. In particular implementations, the original image may be received as input and enhanced and de-noised using image enhancement techniques, and features of the image, including pixel values, spectral features, and texture features, are extracted by the system using computer vision and image processing techniques, training an image quality assessment model. By correlating image features with their corresponding quality levels and problem labels, based on a supervised learning algorithm, such as a support vector machine or Convolutional Neural Network (CNN), with reference to a method of obtaining how accurately to identify and classify quality problems of images, a trained model can be used to detect quality problems, such as blur, noise, and insufficient contrast, present in an input image, while also giving a specific location and severity of each problem. The quality problems and the severity of the images are comprehensively analyzed, comprehensive quality grade evaluation is provided for each image, and the accuracy and the robustness of the system are continuously improved through optimizing and iterating the system, such as introducing more quality problem labels, improving model training strategies and performance indexes.
Referring to fig. 3, a convolutional neural network model is designed, and the steps for training the convolutional neural network model based on the data set are specifically as follows:
designing a framework of a convolutional neural network, and defining a network structure, a convolutional layer, a pooling layer and a full-connection layer to extract image features;
selecting a proper loss function and an optimizer for the convolutional neural network, and performing network training by utilizing the data set after image enhancement processing;
and (5) performing model tuning.
The loss function adopts the difference between the absolute error measurement predicted value and the real quality value, the optimization algorithm adopts an Adam optimizer, and the effect of the training process is adjusted based on the mode of adjusting the learning rate parameter.
The model tuning specifically means that the optimal hyper-parameter configuration is found by adopting a grid search mode, the complexity of the model is controlled by a regularization method comprising L1 regularization and L2 regularization, the fitting problem is relieved, the performance of the model on different data subsets is evaluated by using cross verification, and the optimal model configuration is selected.
Specifically, for image quality assessment, a Convolutional Neural Network (CNN) based architecture is designed in which multiple convolutional, pooling, and fully-connected layer components are included for extracting features from an image. By selecting different convolution kernel sizes, filter numbers, stride and pooling layer size parameters, features of different layers can be extracted, an image dataset with labels is prepared as training data, and in order to optimize a model, a proper loss function is defined to measure the difference between the model output and a real quality value, wherein the loss function comprises absolute error loss and mean square error loss. The Adam optimizer is selected as the optimizer, the learning rate is adaptively adjusted, model parameters are gradually updated in the training process to minimize a loss function, a data enhancement technology is applied to amplify a training set in order to improve the robustness of the model, the diversity of the data set is increased through rotation, scaling and translation operations, and different transformation and viewing angles in the real world are simulated, so that the generalization capability of the model is enhanced. In the aspect of model tuning, a grid search method is adopted to find the optimal super-parameter configuration, such as the convolution kernel size, the filter number and the pooling layer size, and the performance of the model on the verification set is evaluated by trying different super-parameter combinations, and the configuration with the optimal performance is selected. In addition, introducing regularization methods (e.g., L1 regularization, L2 regularization) helps control the complexity of the model and alleviates the over-fitting problem, making the model more prone to choosing a simple weight configuration by adding regularization terms to the loss function, thereby avoiding over-fitting the training data.
Referring to fig. 4, based on the convolutional neural network model, the steps for performing image quality evaluation are specifically:
preprocessing the data of image quality assessment by adopting an image enhancement technology;
loading an image quality evaluation algorithm and a feature extraction method based on computer vision and image processing technology, and automatically detecting and identifying image quality problems by analyzing pixel values, spectrum features and texture features of an image;
performing quality evaluation on the underground coal mine image by adopting a deep learning model;
with reference to the multiple sets of image quality assessment indicators, a composite image quality score is calculated.
The image enhancement technology comprises noise removal, contrast enhancement and sharpening, wherein the noise removal specifically comprises the adoption of a filtering technology, the reduction of noise in an image is based on a median filtering and Gaussian filtering method, the contrast enhancement specifically comprises the adoption of a histogram equalization and contrast stretching method, the contrast and visual effect of the image are improved, and the sharpening specifically comprises the adoption of a Laplace operator to improve the definition of the image.
Image quality assessment indicators include Structural Similarity Index (SSIM), peak signal to noise ratio (PSNR), adaptive contrast enhancement (adaptive contrast enhancement);
the combined image quality scores are calculated by a weighted average method, and weights are distributed based on 40% of Structural Similarity Index (SSIM), 20% of peak signal to noise ratio (PSNR) and 40% of adaptive contrast enhancement (adaptive contrast enhancement).
In particular, the use of a deep learning model for quality assessment of images downhole in a coal mine has several beneficial effects, firstly, the quality of the image can be improved by processing the input image through image enhancement techniques, including noise removal, contrast enhancement and image sharpening. Noise interference in the image can be reduced by the noise removal technology, so that definition and visual quality of the image are improved, contrast of the image can be increased by the contrast enhancement technology, the image is clearer and easier to analyze, details and definition of the image can be improved by the image sharpening technology, and important features are more prominent. Second, the introduction of multiple sets of image quality assessment indicators, including Structural Similarity Index (SSIM), peak signal-to-noise ratio (PSNR), and adaptive contrast enhancement (adaptive contrast enhancement), enables the overall assessment of various aspects of an image, which can quantitatively measure the structure, signal-to-noise ratio, and contrast enhancement effects of the image, thereby providing more comprehensive image quality assessment information. The comprehensive evaluation result for evaluating the underground coal mine image quality can be obtained by calculating the comprehensive image quality score and based on the set weight proportion, and the contribution of each index is synthesized by adopting a weighted average method, so that the importance of each index is ensured to be reasonably considered. The comprehensive score can more accurately reflect the overall quality of the image, provides powerful basis for further analysis and decision, and finally, the trained deep learning model can combine the image enhancement technology and a plurality of groups of image quality evaluation indexes to provide accurate and comprehensive image quality evaluation results. The process can help the coal mine industry evaluate the image quality from the detail level, improve the reliability of image analysis and decision, improve the safety, the production efficiency and the monitoring capability of the coal mine by optimizing related coal mine operation and monitoring tasks and improving related processes and equipment, perform quality evaluation on the underground coal mine image by adopting a deep learning model, and provide more accurate and comprehensive image quality evaluation results by combining an image enhancement technology and a plurality of groups of image quality evaluation indexes, thereby providing valuable information and support for decision and operation of the coal mine industry.
Referring to fig. 5, the performance evaluation of the convolutional neural network model specifically includes the following steps:
adopting a random number generator, taking 80% of data as a training set, 20% of data as a testing set proportion, and dividing the data set into the training set and the testing set;
and performing performance evaluation on the trained model by using an independent test data set, wherein evaluation indexes of the performance evaluation comprise accuracy, recall, precision and F1 score.
Cross-validation specific K-fold cross-validation, dividing the dataset into K folds (folds), using K-1 folds each time as the training set, the remaining 1 fold as the validation set, repeating K times, taking the average performance as a comprehensive assessment of the model.
In particular, the steps of model performance assessment, including data set segmentation, independent testing, and cross-validation, are critical to accurately assessing the performance of a deep learning model and can bring about a series of benefits. Loading a data set containing large-scale coal mine underground image data, randomly shuffling the data set to ensure randomness of division, and calculating the sample numbers of the training set and the test set according to the division proportion (80% of data is used as the training set and 20% of data is used as the test set). And selecting samples from the shuffled data set according to the calculated sample number, dividing the samples into a training set and a testing set, marking the images in the data set, marking the quality grade and specific image quality problem of each image, and providing reliable estimation of the prediction capability of the model in practical application.
In the independent test stage, the performance of the trained model is evaluated by using an independent test data set, and various evaluation indexes including accuracy, recall, precision and F1 score are calculated, wherein the indexes can provide the evaluation of the performance of the model in different aspects and help comprehensively understand the performance of the model on the test set. By loading the test dataset and evaluating using the model, we can compare with the true quality level and problem labels, thereby calculating the accuracy, recall, precision, and F1 score evaluation metrics that can measure the overall performance, problem identification capability, and accuracy of the model. By calculation of these evaluation metrics, a comprehensive understanding of the performance of the model on the test set is available to evaluate the model's effectiveness and make further improvements and optimizations.
Further, cross-validation, such as K-fold cross-validation, is introduced by subjecting the original data set to a shuffling operation to disrupt the order of the data, then dividing the data set into K mutually disjoint subsets, each subset being referred to as a fold, with one of the folds being selected as the validation set and the remaining K-1 folds being the training set for each iteration of cross-validation. In each iteration, model training is performed using data in the training set and model evaluation is performed using data on the validation set. In each iteration, calculating evaluation indexes such as accuracy, recall, precision and F1 score, recording the evaluation indexes of each iteration, and carrying out average operation on the evaluation indexes of K times after K iterations are completed to obtain the comprehensive evaluation result of the model. The indexes can provide average performance performances of the model on different verification sets, the performances of the model on different indexes are analyzed according to comprehensive evaluation results, and the model with the best performance is selected as a final model, so that dependence on a specific data set can be reduced, and generalization capability and stability of the model are better evaluated. In combination, the integration of these model performance evaluation steps provides a systematic and comprehensive way to evaluate the performance of the deep learning model, and through independent testing and cross-validation, multiple performance indexes of the model on different data sets can be obtained, which helps to more accurately understand the performance of the model. The evaluation process is favorable for optimizing the configuration and parameter selection of the model, and improves the prediction capability and generalization capability of the model in practical application, so that the overall effect and reliability of the model are improved.
Working principle: in the whole, the characteristics of the image are extracted through a Convolutional Neural Network (CNN), the information in the image is captured, and the quality of the model to the input image is improved through an image enhancement technology, so that the characteristics of the image are extracted more accurately. The method of comprehensively considering a plurality of evaluation indexes is adopted to comprehensively evaluate the image quality, the method of dividing the data set and independently testing the data set is adopted to perform model training and performance evaluation on the trained model, and the model is verified on different data subsets for a plurality of times through a cross verification method, so that the reliability and generalization capability of model evaluation are further improved, and the deviation caused by improper selection of the data set is reduced. The diversity of the data set is increased through the data enhancement technology, the robustness of the model is improved, and the method is suitable for different sample changes. The model is further optimized through a super-parameter tuning and model regularization method, the problem of over-fitting is relieved, and cross verification is introduced. If the K folds are cross-verified, the performance of the model is more comprehensively evaluated, the data set is divided into K folds, K-1 folds are selected as training sets each time, the rest 1 folds are used as verification sets, the K times are repeated, and the performance of the model on different data sets is comprehensively evaluated according to the average value of the K times of evaluation results. Such a comprehensive evaluation can provide more robust and reliable performance evaluation results, reducing reliance on specific data sets. Specifically, prior to quality assessment of the coal mine downhole image, the image is first processed using image enhancement techniques including noise removal, contrast enhancement and image sharpening steps. Noise in the image can be reduced by a median filtering or Gaussian filtering method, contrast enhancement technology can increase contrast and visual effect of the image by a histogram equalization or contrast stretching method, image sharpening technology can utilize Laplacian to improve definition and detail of the image, a preprocessed image dataset is used for training a deep learning model, the model can adopt a Convolutional Neural Network (CNN) structure, and quality assessment tasks are carried out by learning characteristics of the image. In the training process, a random gradient descent (SGD) optimizer is used for updating parameters of the model, so that the model can adapt to the characteristics and requirements of an underground coal mine image quality evaluation task. Dividing the preprocessed image data set into a training set and a testing set according to the proportion of 80% and 20%, wherein the training set is used for training a model and optimizing parameters, and the testing set is used for evaluating performance. The segmentation mode can ensure that the model is evaluated on an independent data set, so that the performance of the model is known more accurately, the performance of the trained model is evaluated by using an independent test set, and proper evaluation indexes such as accuracy, recall, precision and F1 score are selected according to specific tasks and requirements. And comparing the predicted result of the model on the test set with the real label to calculate the performance evaluation index of the model on the test set.
The present invention is not limited to the above embodiments, and any equivalent embodiments which can be changed or modified by the technical disclosure described above can be applied to other fields, but any simple modification, equivalent changes and modification made to the above embodiments according to the technical matter of the present invention will still fall within the scope of the technical disclosure.
Claims (10)
1. The underground coal mine image quality evaluation method is characterized by comprising the following steps of:
data acquisition, integrating and generating a data set;
designing a convolutional neural network model, and performing model training on the convolutional neural network model based on the data set;
performing an image quality assessment based on the convolutional neural network model;
and performing performance evaluation on the convolutional neural network model.
2. The method of claim 1, wherein the data set comprises coal mine downhole images of various quality levels, the coal mine downhole images of various quality levels comprising common image quality problems including blur, noise, insufficient contrast;
the step of data acquisition and integration generation of the data set comprises the following steps:
collecting and labeling quality grades and problem labels for large-scale coal mine underground image data sets;
the data enhancement technology is applied, including rotation, scaling and translation, so that the diversity of a data set is increased and the robustness of a model is improved;
the data is enhanced by the integrated data enhancement technology, and the data sets are integrated and classified.
3. The method for evaluating the image quality under the coal mine well according to claim 1, wherein the step of designing a convolutional neural network model and performing model training on the convolutional neural network model based on the data set is specifically as follows:
designing a framework of a convolutional neural network, and defining a network structure, a convolutional layer, a pooling layer and a full-connection layer to extract image features;
selecting a proper loss function and an optimizer for the convolutional neural network, and performing network training by utilizing the data set after image enhancement processing;
and (5) performing model tuning.
4. A method of evaluating image quality downhole in coal mine as claimed in claim 3, wherein the loss function uses the difference between the absolute error metric predicted value and the true quality value, the optimization algorithm uses Adam optimizer, and adjusts the effect of the training process based on adjusting the learning rate parameter.
5. A method for evaluating image quality under coal mine as claimed in claim 3, wherein the performing model tuning specifically means searching for optimal super-parameter configuration by using a grid search mode, and the regularization method includes L1 regularization and L2 regularization to control complexity of the model, alleviate the over-fitting problem, evaluate performance of the model on different data subsets by using cross-validation, and select optimal model configuration.
6. The method for evaluating the image quality under the coal mine as claimed in claim 1, wherein the step of performing the image quality evaluation based on the convolutional neural network model specifically comprises:
preprocessing the data of the image quality assessment by adopting an image enhancement technology;
loading an image quality evaluation algorithm and a feature extraction method based on computer vision and image processing technology, and automatically detecting and identifying image quality problems by analyzing pixel values, spectrum features and texture features of an image;
performing quality evaluation on the underground coal mine image by adopting a deep learning model;
with reference to the multiple sets of image quality assessment indicators, a composite image quality score is calculated.
7. The method for evaluating the image quality under the coal mine as claimed in claim 6, wherein the image enhancement technology comprises noise removal, contrast enhancement and sharpening, the noise removal specifically comprises a filtering technology, noise in the image is reduced based on a median filtering and Gaussian filtering method, the contrast enhancement specifically comprises a histogram equalization and contrast stretching method, the contrast and visual effect of the image are improved, and the sharpening specifically comprises a laplace operator for improving the definition of the image.
8. The method for evaluating the image quality under the coal mine as claimed in claim 6, wherein the image quality evaluation index comprises a structural similarity index, a peak signal to noise ratio and an adaptive contrast enhancement;
the combined image quality scores are calculated by a weighted average method, and weights are distributed based on 40% structural similarity index, 20% peak signal to noise ratio and 40% adaptive contrast enhancement.
9. The method for evaluating the image quality under the coal mine well according to claim 1, wherein the step of evaluating the performance of the convolutional neural network model specifically comprises the following steps:
adopting a random number generator, taking 80% of data as a training set, 20% of data as a testing set proportion, and dividing the data set into the training set and the testing set;
performing performance evaluation on the trained model by using an independent test data set, wherein evaluation indexes of the performance evaluation comprise accuracy, recall, precision and F1 score;
and performing cross-validation.
10. The method for evaluating the image quality under the coal mine as claimed in claim 9, wherein the cross-validation is specific K-fold cross-validation, the data set is divided into K folds, each time using K-1 folds as a training set, the remaining 1 folds as a validation set, repeating K times, and taking the average performance as a comprehensive evaluation of the model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311175624.2A CN117197092A (en) | 2023-09-12 | 2023-09-12 | Underground coal mine image quality assessment method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311175624.2A CN117197092A (en) | 2023-09-12 | 2023-09-12 | Underground coal mine image quality assessment method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117197092A true CN117197092A (en) | 2023-12-08 |
Family
ID=88995697
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311175624.2A Pending CN117197092A (en) | 2023-09-12 | 2023-09-12 | Underground coal mine image quality assessment method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117197092A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117593597A (en) * | 2024-01-19 | 2024-02-23 | 山东省国土测绘院 | Automatic classification method and system for topographic images |
| CN118037742A (en) * | 2024-04-15 | 2024-05-14 | 山东中联晶智信息科技有限公司 | Multimedia image quality evaluation system and method based on scene recognition |
| CN118097386A (en) * | 2024-04-24 | 2024-05-28 | 四川众力佳华信息技术有限公司 | Ammeter photo credibility verification method |
-
2023
- 2023-09-12 CN CN202311175624.2A patent/CN117197092A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117593597A (en) * | 2024-01-19 | 2024-02-23 | 山东省国土测绘院 | Automatic classification method and system for topographic images |
| CN117593597B (en) * | 2024-01-19 | 2024-03-22 | 山东省国土测绘院 | Automatic classification method and system for topographic images |
| CN118037742A (en) * | 2024-04-15 | 2024-05-14 | 山东中联晶智信息科技有限公司 | Multimedia image quality evaluation system and method based on scene recognition |
| CN118097386A (en) * | 2024-04-24 | 2024-05-28 | 四川众力佳华信息技术有限公司 | Ammeter photo credibility verification method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114549522B (en) | Textile quality detection method based on target detection | |
| CN110349126B (en) | Convolutional neural network-based marked steel plate surface defect detection method | |
| CN117197092A (en) | Underground coal mine image quality assessment method | |
| CN107657279B (en) | Remote sensing target detection method based on small amount of samples | |
| CN116758491B (en) | Printing monitoring image analysis method and system applied to 3D printing | |
| CN113706434B (en) | Post-processing method for chest enhancement CT image based on deep learning | |
| CN110689524B (en) | No-reference online image definition evaluation method and system | |
| CN109785301A (en) | A kind of rail corrugation automatic identifying method and appraisal procedure based on image procossing | |
| Zhang et al. | A method for defect detection of yarn-dyed fabric based on frequency domain filtering and similarity measurement | |
| CN110033434A (en) | A kind of detection method of surface flaw based on texture conspicuousness | |
| Liang et al. | Automatic defect detection of texture surface with an efficient texture removal network | |
| CN115294377A (en) | System and method for identifying road cracks | |
| CN116434206A (en) | Cotton quality character detection method based on machine vision technology | |
| Mollajan et al. | Improving pore type identification from thin section images using an integrated fuzzy fusion of multiple classifiers | |
| CN111783616A (en) | Data-driven self-learning-based nondestructive testing method | |
| Dong et al. | Fusing multilevel deep features for fabric defect detection based NTV-RPCA | |
| CN114897825A (en) | Solid wood floor sorting method and system based on computer vision | |
| CN116777917A (en) | Defect detection method and system for optical cable production | |
| CN117314826A (en) | Performance detection method of display screen | |
| CN116298190A (en) | Screening and determining method for geochemical investigation index elements | |
| CN115640546A (en) | Lithology identification method based on fusion of image and feature information | |
| CN114998876A (en) | Sea-land transition phase shale streak layer structure identification method based on rock slice image | |
| CN112508946A (en) | Cable tunnel abnormity detection method based on antagonistic neural network | |
| CN116958128B (en) | Medical image automatic positioning method based on deep learning | |
| CN118097313B (en) | Hyperspectral image classification method based on frequency domain active learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |