Abstract
Selecting reasonable blasting parameters of ore and rock is an important measure to achieve good blasting effect. In the mining process, rock fragmentation is an important index to evaluate the blasting effect, which directly affects the technical scheme, equipment selection, economic effect, and other issues of the mine and even seriously threatens the sustainable safety production of the mine. With the rapid development of information technology, the development of computer intelligent image recognition technology is becoming more and more perfect, and its role is becoming more and more important. Based on the neural network method, this paper studies the computer intelligent image recognition technology. In this paper, the GA-BP network image recognition model is established by combining genetic algorithm with BP algorithm and analyzing the principles of intelligent image recognition, image pattern recognition, and BP neural network learning algorithm. On the basis of experimental analysis, the average accuracy of prediction can reach 67.4%. For the efficiency analysis of computer mathematical analysis, it will generally reach 64.3%. In this paper, taking the lump rate and blasting cost as the optimization objective function, the comparison and selection of multiple schemes of production blasting design are carried out, which provides quantitative decision-making basis for the rational selection of production blasting design parameters.
1. Introduction
The rock stripping and ore recovery of large-scale open-pit iron mines mostly use deep hole blasting, which has a high degree of blasting mechanization, fast construction speed, and relatively concentrated blasting [1]. In order to accurately predict the ore-rock boundary after blasting, guide shoveling equipment to accurately locate high-grade ore, and minimize ore dilution caused by waste rock mixing during blasting, it is necessary to explore and study the movement of broken ore rock during blasting [2]. With the development of computer application technology, blasting workers have applied computer simulation technology to blasting and made remarkable progress in blasting effect prediction and optimization design. Many open-pit mines in China have also achieved good results by using this technology in blasting production [3]. Image analysis is a new method for measuring the lumpiness of ore rock which has been gradually developed in recent years. The rock fragmentation and its distribution are affected by many factors such as rock properties, explosive properties, mesh parameters, and explosive unit consumption, which belong to a multifactor comprehensive prediction and analysis problem.
Over the years, blasting experts at home and abroad have done a lot of research and experiments on the internal relationship and objective laws between them [4, 5]. With the development of blasting theory and computer technology, a number of effective theoretical achievements proved by production practice have been obtained, and the blasting effect can be reliably predicted and analyzed [6]. The efficiency and accuracy of computer intelligent image recognition are much higher than that of human sensory object recognition. Computer intelligent fusion image recognition technology can intelligently process image information [7]. Through such intelligent processing, after the information is processed, the anti-interference ability is stronger, and the image information can be easily transmitted and stored. Image data acquisition can be defined as the transmission of scenes, images, photos, etc. through digital cameras, scanners, fax machines, and other equipment through a suitable sensor, finally transformed into computer processing forms, which can also be understood as transforming physical forms into measurement forms [8]. Image data processing can be divided into three parts: feature selection, feature extraction, and preprocessing. Preprocessing is a variety of image processing technologies. After preprocessing, the image quality has been greatly improved, which can solve the so-called digestion phenomenon [9, 10]. Feature extraction can be understood as the extraction of digital values and basic elements of image features.
At present, the complexity of network system is reflected in the following aspects: first, the structure of network is extremely complex, and there is no unified concept of the connection between network nodes so far. Second, the network is constantly evolving, the nodes of the network are also increasing, and the connections between nodes are also growing. In addition, there are many diversities between connections. Third, the complexity of network dynamics is very complex. Each node can be a nonlinear system. Its behavior has the characteristics of nonlinear dynamics, such as chaos and bifurcation, and is constantly changing [11, 12]. The two important steps of quantification and regional segmentation were studied, and a blasting rock distribution model with strong applicability and high processing efficiency was established, and a software package for automatic determination of ore block degree was developed [13–15]. However, the above research has not solved the corresponding problems well, so this paper puts forward the following innovations: ① Combining genetic algorithm and BP algorithm, a computer image recognition algorithm based on GA-BP depth neural network is proposed. Compared with the BP neural network method, the proposed GA-BP deep neural network algorithm overcomes the shortcomings of slow convergence speed and long training time. ② Kuznetsov and Rosin–Rammler mathematical models are used as mathematical models to predict the average fragmentation of blasting rock and its distribution, and the fragmentation rate and blasting cost are used as optimization objective functions to compare and select multiple schemes of production blasting design, which provides quantitative decision-making basis for reasonable selection of production blasting design parameters.
The chapters of this paper are arranged as follows: the first chapter is the introduction, which discusses the background and significance of the topic selection and expounds the innovation of the article. The second chapter of this paper mainly combines the domestic and foreign research results in the field of mine blasting quality prediction and analysis based on GA-BP neural network and proposes innovative results and research ideas of this paper. The third chapter of this paper is the method part, which deeply discusses the application and principle of related algorithms, and puts forward the prediction, analysis, optimization, and application model of ore and rock blasting quality based on the previous research results and the innovation of this paper. The fourth chapter mainly discusses the experimental part of the application of the algorithm. Through the experimental results, the model is established on the basis of sorting out the data. The fifth chapter is the summary part, which summarizes the research results of this paper.
2. Related Work
Dhekne et al. believe that the measurement of the size of the blasted rock blocks is still few, and there is no accepted method; one of the main reasons is the complexity and irregularity of the blasted rock accumulation nature making image processing extremely difficult [16]. Xin et al. put forward that the degree of blasting fragmentation of ore and rock not only determining the efficiency and cost of blasting work itself, but also directly affecting the production efficiency and cost of subsequent processes, such as loading, transportation, and initial fragmentation, thus also affecting the total cost of mining production [17]. The research of Ye et al. shows that rock drilling and blasting is an essential step in the mining process. The blasting effect directly affects the smooth progress of subsequent shovel loading work. At the same time, blasting is a complex process involving many factors, with instantaneous fuzziness and uncertainty [18]. Riu et al. believe that image pattern recognition can be defined as the recognition and processing of pattern information such as pictures, images, and texts by machines, and it can also be understood that image recognition can replace people to automatically process and classify certain information with identification. Image recognition mainly includes three parts: image pattern classification, image feature extraction, and image preprocessing. The preprocessing of the image includes image enhancement, smoothing, segmentation, and edge detection, which can transform image recognition into a segmentation processing mode [19]. Zhang et al. pointed out that by integrating image recognition technology with computer intelligence, the computer image recognition technology has been completely innovated. For example, the sensor five-color low illumination processing technology can improve the monitoring equipment and camera resolution, thus improving the image recognition degree. Using high sensitivity technology, the picture can be taken very clearly in dark light [20]. Sayevand et al. believe that the information security model is used to accurately describe the security policy of the information system and use formal or informal methods to describe the security policy of the information system. The information security model is based on the architecture, and the information security architecture is based on various information security technologies, methods, and mechanisms. It is an organic combination of subset elements of various mechanisms [21]. Li et al. believe that cluster is the concept of agglomeration degree. For example, there may always be a circle of friends or a circle of acquaintances in a social network, and each of them may know other members. Therefore, the meaning of clustering is the degree of network's grouping; this is a tendency of network cohesion. The concept of connected group reflects the distribution and interconnection of each clustered small network in a large network [22]. Faradonbeh and Monjezi proposed that the ratio of hole spacing to row spacing, hole depth to row spacing, row spacing to hole diameter, hole plugging length to row spacing ratio, explosive unit consumption, in-situ rock fragmentation, and rock elastic modulus should be taken as input parameters to predict blasting fragmentation [23]. The research of Onyelowe et al. shows that the blast hole row spacing, hole bottom spacing, and primary explosive unit consumption are the most important blasting parameters of ore and rock. The inherent properties of ore and rock, including bulk density, elastic modulus, tensile strength, compressive strength, friction angle, and cohesive force, are the main factors affecting the blasting parameters of ore and rock. Therefore, there must be a certain functional correlation between the ideal blasting parameters and the main factors [24]. Tao et al. showed in the experiment that taking bulk density, elastic modulus, tensile strength, compressive strength, friction angle, and cohesive force as input factors, the blasting hole row spacing, hole bottom spacing, and the unit consumption of primary explosive are the output factors and the middle hidden layer is set to form a 3-layer network structure. The size of the training sample set is extremely important. Too little learning is of little significance, and it affects the training speed or even fails to converge to the predetermined accuracy. It is appropriate to take 6 to 25 [25]. Nelsonk thinks that the definition of blasting fragmentation is the size of fragments formed by ore and rock after blasting. Generally, the following methods are used to describe the fragmentation degree of ore and rock blasting: fragmentation distribution function, average blockiness, and qualified blockiness and then study the blockiness problem from a quantitative point of view [26]. Hu, Wu, and Zhao proposed that if the blasting fragmentation is too large, it needs secondary crushing, so as to reduce the production capacity of ore transportation, produce dust and toxic gas, and pollute the working environment. There are serious safety problems when dealing with large blocks squeezing and blocking the chute or funnel, and breaking large blocks at the leakage gate will often lead to accidents such as gate collapse and cable damage [27]. Jemmali studied the nonlinear complex relationship between blasting parameters and their main influencing factors [28]. Han thinks that it is adaptive to the establishment of the relationship model between the blasting parameters and the main influencing factors and that the optimization process of blasting parameters does not need to establish mathematical equations, which has the characteristics of self-adaptability, learning ability, fault tolerance, and robustness and can avoid the disadvantages of the traditional method for determining the blasting parameters of ore and rock [29]. Jhaveri designed the GA-BP network environment quality evaluation model structure. During training, the initial weights of the network were optimized by genetic algorithm, and then the BP learning algorithm was used for training after the search scope was narrowed. Finally, the generalization ability of the network was used to predict the test samples [30].
Based on the research of the above-mentioned related work, this paper determines the positive role in the field of mine blasting quality prediction and analysis method based on GA-BP neural network. The data uses GA-BP neural network analysis to conduct in-depth analysis and research, make more effective use of data, mine valuable knowledge hidden behind the data, and discover and find potential problems that affect the prediction and analysis of rock blasting quality.
3. Methodology
3.1. Research and Analysis of Related Theories
3.1.1. The Basic Principle and Model of Artificial Neural Network
Neural network is a large-scale nonlinear dynamic and parallel distributed information processing system, which is made up of many simple processing units (i.e., neurons or nodes) connected with each other, drawing on the characteristics and structure of human brain. It has a series of characteristics, such as huge parallelism, structural variability, high nonlinearity, self-organization, and self-learning.
The Internet has distinct complex system characteristics. From the perspective of openness, the users of the Internet are people, and the realization of Internet technical specifications and various platforms also depend on people's operations. Therefore, the Internet must interact with the surrounding environment, and social environment and people play an important role in the development of the Internet. From the perspective of hierarchy, the Internet implements a hierarchical addressing scheme through domain names and addresses. In different domain name system levels of the Internet, they have their own network control center and network information center, which are responsible for the monitoring and control of network operation status within their jurisdiction. Figure 1 shows a general description of neurons.
The neural network is composed of many simple processing units, which are connected by variable weights to form a parallel distributed system. In addition, the basic processing unit of the neural network is a neuron, and its structure is shown in Figure 1. In the figure, is the input signal, represents the connection weight from the -th neuron to the -th neuron, and the threshold value of the -th neuron is . Let be the external input signal and be the output signal. In this model, the transformation of the neuron can be described as
The nonlinear function used here can be a step function, a piecewise function, and a sigmoid-type function. For any neuron, there is a corresponding output, and the output is transmitted to the processing unit connected to it through the connected weight. The output signal is directly dependent on the state or excitation value of the processing unit. The function of processing unit is represented by the output transformation function . If we use to define the output of neurons at moment, then its general expression is
Or, express in vector form:
Among them, is on the neural network as an output vector, and is defined as a state vector with its corresponding function of each component. This function is usually on the interval and is bounded.
The BP neural network algorithm can be regarded as a gradient descent method, which constantly adjusts the threshold value, weights, and other parameters in the calculation process to achieve the minimum mean square error of the actual output value and the expected value. The calculation method of BP algorithm index function iswhere can be expressed as
The methods to minimize the objective function can be divided into two types, the most classic is batch processing and one-by-one processing. Processing one by one means that the samples are input in sequence, and the connection weights are adjusted each time a different sample is input. Batch processing can be understood as waiting for all samples to be input and then iteratively calculating the total error. The flow chart of BP algorithm is shown in Figure 2.
3.1.2. Research on Algorithm Based on GA-BP Neural Network
Although the BP algorithm has excellent self-learning, it adopts the gradient descent method for calculation, the algorithm convergence speed is relatively slow, it is easy to fall into the local minimum, and the training time is relatively long. This paper combines GA and BP algorithm based on genetic algorithm to overcome this disadvantage. GA-BP algorithm can carry out global search, so the disadvantage of immature convergence can be avoided. Genetic algorithm can converge the global optimal solution, which has strong robustness. The combination of neural network and genetic algorithm makes the nonlinear mapping ability of neural network fully exerted and has fast convergence speed and strong learning ability. In the fitness function evaluation, the evaluation method adopts the mean square error; then the mathematical model of the GA-BP algorithm is
in which , where is the weight matrix of input layer and output layer; is the threshold matrix of input layer and output layer; is the number of layers of input layer, hidden layer, and output layer; represents the sample number of the hidden layer; and is the sample error. Figure 3 shows the basic algorithm flow chart of GA-BP neural network.
The ore-rock structure surface mainly includes joints, fissures, bedding, fractures, and folds. When the direction of action is the same, the explosive gas intrudes into the structure surface to open the crack surface, resulting in a decrease in the pressure in the blast hole, thus affecting the fracture of the rock. When they are perpendicular to the direction of blasting action, it is favorable for the propagation and reflection of blast stress waves. Table 1 shows the influence of hole network parameters on blasting quality. Table 2 shows the influence of explosive unit consumption on blasting quality.
In addition, the structural surface will also affect the propagation of the blast stress wave, thereby affecting the blasting effect. Blasting parameters refer to the parameters that embody and exactly explain different blasting methods and schemes and adopt various indicators corresponding to drilling and blasting technology. The main blasting parameters, such as charge length, charge diameter and density, interval charge, and explosive number, are used to explain the specific form of charge. The explosion parameters, such as the number of initiation stages, the time difference of each stage, the length of detonation transmission, and the blasting range, directly affect the important indexes such as explosion energy, energy utilization rate, and rock crushing mechanism, thus determining the crushing effect of ore and rock. In this paper, Kuznetsov and Rosin–Rammler numerical simulation are used as the mathematical model for predicting the average fragmentation and distribution of blasting ore and rock. Taking the chunk rate and blasting cost as the optimization objective function, the multischeme comparison and selection of production blasting design are carried out, which provides a quantitative decision-making basis for the reasonable selection of production blasting design parameters.
The mathematical model for predicting the average fragmentation of Kuznetsov ore rock isin which represents the average fragmentation of ore and rock, represents the rock characteristic coefficient, represents the rock volume of each blast hole, represents the single hole charge, and is the explosive weight power.
The prediction model of Rosin–Rammler ore rock fragmentation distribution is established based on this requirement, which can better describe the distribution law of ore rock fragmentation.
Among them, represents the mass percentage of the material when the size of the sieve is , represents the size of the sieve, represents the characteristic lumpiness, and represents the index that determines the shape of the lumpiness distribution curve. And the larger the value of , the narrower the distribution range of rock lumpiness, which means that there are fewer fine ores and ultra-large lumps, and vice versa.
Because the genetic algorithm takes the maximum value of the objective function as its fitness function in the process of optimization, the fitness function is defined as
At this point, the optimization equation can be changed to
First, the basic solution space is encoded, and the code string generated by the encoding is composed of two parts, the control code and the weight coefficient code. The control code is mainly used to control the number of hidden nodes. It is a string consisting of 0–1, where 0 means not connected and 1 means connected. The string length l1 can be determined by 0.5 to 1.5 times of the number of input nodes. The coefficient of weight is mainly used to control the connection right of the network and adopt the coding of floating-point numbers, with a string length of .
4. Result Analysis and Discussion
Based on the above research analysis, this paper conducts the following experiments to illustrate the data. The computer simulation research on the physical process of rock blasting in open-pit mines has the dual characteristics of strong theory and strong practice. It is its aim to apply the theoretical research results of rock blasting to production and practice in order to realize the optimization of open-pit blasting. However, due to the complexity of the rock blasting process and the immaturity of the rock blasting theory, it is impossible for us to include all the factors when constructing the mathematical model, so a mathematical model can only prominently reflect that it plays a major role in the rock blasting process. Practice has proved that the crushing effect of stress wave is indeed basic and main for hard brittle rock. For soft rock, the elastic rheological theory of outburst explosive gas should be considered to construct the mathematical model. For this reason, this paper tries to conduct practical analysis from three aspects: sample prediction error, computer mathematical analysis efficiency, and performance comparison between GA-BP algorithm and BP algorithm. Figures 4 and 5 are the experimental analysis diagrams of training sets A and B on sample prediction error and computer mathematical analysis efficiency.
Through the above experimental analysis, it can be known that the prediction of blasting quality of ore and rock fluctuates by more than 40%, which will also ensure that the analysis based on GA-BP neural network is in a relatively reasonable and credible interval. Through observation, it can be found that the performance of the two sample sets basically maintains the same trend in the experiment, which also shows that the algorithm designed in this paper is typical in general. Overall, the average accuracy of prediction can reach 67.4%. In Figure 5, it can be found that the computer processing has regular fluctuations, which also reflects the characteristics of the computer's data processing. For different sample sets, the processing capability of the computer appears to be different. This is also because there are different interference items in each different sample set, and the computer has different influences on the interference items, so the final results are also different. However, for the efficiency analysis of computer mathematical analysis, it usually reaches 64.3%.
In order to prevent the loss of accuracy caused by different dimensions and orders of magnitude of training parameters and to reduce the impact of the error of the maximum and minimum values on the whole data set, the two groups of data are standardized respectively. Table 3 shows the data after standardized processing.
In general, the more hidden layers, the better the performance of GA-BP. However, it may lead to too long training time or overfitting phenomenon, so it is extremely important to choose the appropriate number of hidden layers. At present, there is no suitable analytical formula to determine the number of hidden layers. The usual way is to estimate the number of hidden layers according to the empirical formula or to select the appropriate number of hidden layers according to personal experience. Figure 6 shows the performance comparison and analysis of GA-BP algorithm and BP algorithm.
On the whole, GA-BP algorithm model is better than single BP algorithm model. Especially in the overall structure, the GA-BP algorithm is more stable, which will greatly enhance the accuracy of prediction and the effect of analysis, especially for a single BP neural network algorithm, which will basically be in a larger increase or decrease trend. In the concrete operation, the GA-BP algorithm structure converges faster than the BP algorithm structure.
5. Conclusions
In this paper, a computer intelligent image recognition algorithm based on GA and BP neural network is proposed. By analyzing the principle and pattern of image recognition, based on the traditional BP neural network learning algorithm, a GA and BP neural network image recognition model is established, the algorithm flow of the algorithm is deeply analyzed, and the effectiveness of the algorithm is verified by experiments. The theoretical research of blasting block degree control technology is one of the important tasks at this stage. The research fields of rock blasting fracture theory, blasting technology, and physical and mechanical properties of rock are the basic theories of rock blasting fragmentation. Based on the essential characteristics of gray scale distribution of rock blocks in ore and rock images, the double threshold bright spot expansion and decline cycle processing technology is adopted to realize the efficient computer automatic identification of rock block boundaries. From the study of the distribution law of rock fragmentation after blasting, the quantitative relationship between rock, explosive, hole network geometry and process parameters and fragmentation distribution parameters is explored theoretically and experimentally. Because it avoids the complex and mature physical process of rock blasting failure, it is likely to be a shortcut to seek blasting optimization. On the basis of experimental analysis, it is concluded that the average accuracy of prediction can reach 67.4%. For computer mathematical analysis efficiency analysis, it will reach 64.3% in general. As BP neural network is a local optimization algorithm based on gradient descent principle, it has some shortcomings, such as slow convergence speed, easiness to fall into local minimum, weak global search ability, difficulty to determine the network structure, and poor generalization ability. The optimized parameters need to be revised according to the actual production, so as to be consistent with the predicted results.
Data Availability
The data used to support the findings of this study are included within the article.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Authors’ Contributions
Conceptualization and methodology are formulated by Jianyang Yu; software is brought by Shijie Ren; and validation is done by Shijie Ren and Jianyang Yu. Formal analysis and data curation are carried out by Jianyang Yu. Writing (original draft preparation as well as review and editing) is done by Jianyang Yu. Visualization is performed by Shijie Ren. All authors have read and agreed to the published version of the manuscript.