Coal mine safety is crucial to the healthy and sustainable development of the coal industry, and coal mine flood is a major hidden danger of coal mine accidents. Therefore, the processing of coal mine water source data is of great significance to prevent mine water inrush accidents. In this experiment, the water source data were obtained by laser induced fluorescence technology with the assistance of laser. The water sample data information was preprocessed by standard normal variable transformation (SNV) and multiple scattering correction (MSC), and then the principal component analysis (PCA) was used to reduce the dimension of the data and ensure the information characteristics of the original data unchanged. In order to identify the water inrush type of coal mine water source, the sparrow search algorithm (SSA) is used to optimize the BP neural network in this study. This is because the SSA algorithm has the advantages of strong optimization ability and fast convergence rate compared with particle swarm optimization (PSO) and other optimization algorithms. Experiments show that under the premise of SNV pretreatment, the R ² of SSA-BP model is infinitely close to 1, MRE is 0.0017, RMSE is 0.0001, the R ² of PSO-BP model is 0.9995, MRE is 0.0026, RMSE is 0.0019, the R ² of BP model is 0.9983, MRE is 0.0140, RMSE is 0.0075. Therefore, SSA-BP model is more suitable for the classification of coal mine water sources.

Accurate identification of coal and gangue is an important prerequisite for the effective separation of coal and gangue. The application of imaging technology combined with image processing steps (like enhancement, feature extraction, etc.) and classifier is used to identify coal and gangue, which effectively avoids the shortcomings of traditional methods (radiation, pollution, etc.). However, ordinary image detection is greatly influenced by environmental factors such as light, dust and so on. Multispectral imaging technology, as a new generation of optical non-destructive testing technology, is less affected by illumination, so we propose a new solution for the recognition of coal and gangue by using multispectral imaging. Firstly, we respectively tested the classification performance of different image feature extraction methods under GS-SVM, GA-SVM, and PSO-SVM classifiers, and selected the best feature extraction method is LBP. And then, we compared the classification effects under different wavelengths and found that the ninth wavelength works best. That is, the difference in imaging between coal and gangue at 773.776 nm is greatest. Finally, the performance of the proposed model for the identification of coal and gangue was carried out. And the highest classification accuracy can be obtained by using GS-SVM as the classifier, at which point, c=8, g=0.17678. The results show that multispectral imaging technology can be used for the identification of coal and gangue, and the prediction accuracy of the model combined with LBP feature extraction and GS-SVM can reach 96.25% (77/80). The conclusions could provide reference evidence for the intelligent dry selection in coal preparation plants and underground coal mine.

Aiming at the problems of long detection time and low detection accuracy in the existing coal gangue recognition, this paper proposes a method to collect the multispectral images of coal gangue using spectral technology and match with the improved YOLOv5s (You Only Look Once Version-5s) neural network model to apply it to coal gangue target recognition and detection, which can effectively reduce the detection time and improve the detection accuracy and recognition effect of coal gangue. In order to take the coverage area, center point distance and aspect ratio into account at the same time, the improved YOLOv5s neural network replaces the original GIou Loss loss function with CIou Loss loss function. At the same time, DIou NMS replaces the original NMS, which can effectively detect overlapping targets and small targets. In the experiment, 490 sets of multispectral data were obtained through the multispectral data acquisition system. Using the random forest algorithm and the correlation analysis of bands, the spectral images of the sixth, twelfth and eighteenth bands from twenty-five bands were selected to form a pseudo RGB image. A total of 974 original sample images of coal and gangue were obtained. Through two image noise reduction methods, namely, Gaussian filtering algorithm and non-local average noise reduction, 1948 images of coal gangue were obtained after preprocessing the dataset. This was divided into a training set and test set according to an 8:2 ratio and trained in the original YOLOv5s neural network, improved YOLOv5s neural network and SSD neural network. By identifying and detecting the three neural network models obtained after training, the results can be obtained, the loss value of the improved YOLOv5s neural network model is smaller than the original YOLOv5s neural network and SSD neural network, the recall rate is closer to 1 than the original YOLOv5s neural network and SSD neural network, the detection time is the shortest, the recall rate is 100% and the average detection accuracy of coal and gangue is the highest. The average precision of the training set is increased to 0.995, which shows that the improved YOLOv5s neural network has a better effect on the detection and recognition of coal gangue. The detection accuracy of the improved YOLOv5s neural network model test set is increased from 0.73 to 0.98, and all overlapping targets can also be accurately detected without false detection or missed detection. At the same time, the size of the improved YOLOv5s neural network model after training is reduced by 0.8 MB, which is conducive to hardware transplantation.

Mine water inrush disaster can quickly cause significant economic losses and casualties because of its strong concealing and rapid burst speed. Quickly identifying the source of mine water inrush is of great practical significance. Compared with the traditional hydrochemical analysis method, the laser-induced fluorescence (LIF) technology has fast reaction speed, high sensitivity, and strong stability, which makes up for the shortcomings of the traditional method. As an integrated algorithm, random forest (RF) has the advantage of high accuracy. A combination of LIF technology and RF algorithm is proposed to identify mine water inrush source rapidly. The experimental samples were collected from a coal mine in Hainan City, Anhui Province, and a total of 525 sets of water samples to be tested for experiments by mixing goaf water and sandstone water into A-G7 species according to different proportions. Moving average smoothing (MA), Savitzky-Golay Smoothing (SG), First derivative (FD), and Second derivative (SD) methods are used to preprocess the original spectral data to reduce the noise and interference information existing in the original spectral data. By comparison, the Moving average smoothing (MA) method has high classification accuracy and is the final method for noise reduction. Then, the RF algorithm is used to delete the less critical spectrum after noise reduction and select the characteristic wavelength with the minimum classification error of 0. Finally, SVM, PCA-SVM, MA-SVM, MA-PCA-SVM, and MA-RF algorithm recognition models were established, respectively. Comparing the prediction accuracy of the test set, the accuracy of the MA-RF algorithm in the five groups of models reached 100%, which can quickly and accurately predict mine water inrush.

There is a great threat to the production safety of coal mines caused by mine water disasters. Traditional identification methods are not adapted to the efficiency of today’s coal mining and do not offer the advantage of accurate detection in real-time. In this study, the Mayfly Algorithm (MA) was used to optimize the Long Short-Term Memory (LSTM) network, combined with laser-induced fluorescence technology, to apply it to the identification of mine water sources for the prevention of mine water disasters and post-disaster relief work. Taking sandstone water and goaf water as the original samples, five mixed water samples were also prepared by mixing the sandstone water and goaf water in different proportions, giving a total of seven water samples to be tested. Laser-induced fluorescence technology was used to obtain the fluorescence spectral data of water samples, and then the Linear Discriminant Analysis (LDA) dimensionality reduction algorithm and the Principal Component Analysis (PCA) dimensionality reduction algorithm were used to reduce the dimensions of the original spectral data. Then, three architectures, including LSTM, GA-LSTM (optimization of the LSTM by genetic algorithm) and MA-LSTM were designed to identify mine water sources. Finally, from the results’ analysis, MA-LSTM performs best in many aspects after PCA dimensionality reduction and has the best identification effect. These results supported the feasibility of the novel method.