Research On Super-resolution Reconstruction Of Mine Image Based On Online Dictionary Learning

At present,the coal industry is paying more and more attention to the development of mining concepts with "intelligence and safety" as the core.Obtaining high-definition underground video images has become one of the necessary conditions for achieving precise mining in coal mines.The higher the resolution of the underground images acquired means that The higher the amount of data included,this will help coal mine workers get more effective information and play an important role in achieving the safety and intelligence of coal mining.However,the underground environment is special,low illumination,noise interference and other phenomena occur from time to time.Due to the limitation of hardware equipment,the acquired mine images are often of low resolution,which is prone to image blur and edge information loss,and it is difficult to achieve satisfactory visual effects.The super-resolution reconstruction technology of the image can effectively improve the resolution of the mine image and make up for the lost high-frequency detail information.It has important applications in the fields of intelligent coal mining and underground unmanned work.This paper mainly studies the super-resolution reconstruction of mine images under two conditions.In the framework of the classic sparse representation model,combined with dictionary learning methods,two online super-resolution reconstruction methods for mine images based on online dictionary learning are proposed.The main work And innovation is as follows:(1)Under the condition of low illuminance,in view of the problem that the mine image is prone to blur artifacts,the image super-resolution reconstruction algorithm is improved.The shortcomings of the traditional algorithm can not meet the needs of mine image reconstruction.This paper proposes a parameter-adaptive online dictionary learning mine image super-resolution reconstruction algorithm.Based on the classic learning-based method framework,online dictionary learning(ODL)is introduced for dictionary training,and a complete dictionary is obtained.Effectively reduce the dependence on fixed training samples and improve the accuracy of dictionary learning.The reconstruction process of the input image is based on the sparse representation theory.By improving the adaptive method of the parameters,the regular items in the reconstruction phase can be flexibly adjusted to ensure that the parameters can be adaptively determined according to the characteristics of each pixel block to obtain the best Regularization parameters.In order to verify the rationality and advancedness of the proposed algorithm,this algorithm is first applied to the international standard image data set,and the results are compared with other algorithms to analyze the experimental results and data.After comparing and optimizing the algorithm experiments,the algorithm is finally tested on the mine image with fuzzy artifacts under low light conditions,to verify the effectiveness and pertinence of the algorithm in mine image reconstruction.The subjective results and objective evaluation of the experiment prove that the proposed algorithm has a better effect on the reconstruction of the mine image under low illumination conditions under different enlargement factors,which is better than the processing result of the traditional comparison algorithm,which can overcome the mine image reconstruction process.The phenomenon of blurring and visual artifacts.(2)Under noisy conditions,the edge signal of the mine image is seriously lost.For this problem,the reason why the traditional algorithm cannot effectively reconstruct its edge details is analyzed.Combining the characteristics of the image block,an online dictionary with non-local self-similarity is proposed.Learning the super-resolution reconstruction algorithm of mine images.Edge fusion is performed in the preprocessing stage;non-local self-similarity regular terms are introduced into the sparse representation model of mine images.In the dictionary training phase,the sample set is similarly clustered and partitioned by the K-means algorithm,and then the ODL algorithm is used to update the dictionary block for each class partition;in the image reconstruction process,non-local self-similar priors are used After inputting the LR image to extract its own high-frequency signal,after achieving the selection of the best dictionary,it participates in the iterative update of the dictionary again.In this way,both the external dictionary and the input information of the mine image can be used,so that the dictionary used for reconstruction can make up for the lost edge information,and adapt to the structural characteristics of each low-resolution image block.Similarly,a standard image set is selected for comparative experiments to verify the anti-noise performance of the proposed algorithm.After the comparison and analysis of the algorithms,finally,in view of the problem that the edge information of the mine image is easy to be lost under noise conditions,an experimental test was performed on the image captured by the underground video.By comparing with the traditional algorithm,the result data obtained by the algorithm experiment has obvious quality improvement from subjective or objective evaluation,which can more clearly restore the edge texture details of the mine image,improve the noise robustness,and prove its suitability Super-resolution reconstruction processing of mine image edges under noisy environment.There are 41 figures,15 tables and 96 references in this paper.