CN115953731A - Intelligent coal flow monitoring data analysis method for improving CNN algorithm model - Google Patents
Intelligent coal flow monitoring data analysis method for improving CNN algorithm model Download PDFInfo
- Publication number
- CN115953731A CN115953731A CN202211605959.9A CN202211605959A CN115953731A CN 115953731 A CN115953731 A CN 115953731A CN 202211605959 A CN202211605959 A CN 202211605959A CN 115953731 A CN115953731 A CN 115953731A
- Authority
- CN
- China
- Prior art keywords
- coal flow
- flow monitoring
- image
- data
- algorithm model
- 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
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Abstract
The invention discloses an intelligent coal flow monitoring data analysis method for improving a CNN algorithm model, which is applied to the field of coal flow monitoring or information processing; the technical problem to be solved is analysis of coal flow monitoring data, and the adopted technical scheme is an intelligent coal flow monitoring data analysis method for improving a CNN algorithm model, which comprises the following steps that (S1) a dome camera in a monitoring area acquires coal flow monitoring data by adopting a machine vision technology; (S2) denoising and eliminating the collected coal flow monitoring image by an improved dark channel prior filtering method; (S3) the coal flow monitoring image subjected to denoising processing is transmitted to a data storage center through a network in an indirect mode, and an improved CNN algorithm model is adopted to carry out data analysis on the coal flow monitoring image; (S4) transmitting the abnormal analysis result to a monitoring alarm module, and displaying the abnormal analysis result through an OLED display; the invention can reduce the interference of noise to the coal flow monitoring image and improve the accuracy of analyzing the abnormal condition of the coal flow monitoring image.
Description
Technical Field
The invention relates to the field of coal flow monitoring or information processing, in particular to an intelligent coal flow monitoring data analysis method for improving a CNN algorithm model;
background
With the continuous improvement of the mechanization degree of coal production enterprises, the conveying belt becomes an indispensable device in the production process. The idle running rate of the belt is too high in the coal mining process due to various reasons, the idle running rate is too high, electric energy is wasted, the electric charge of the belt occupies a large proportion of the electric consumption in coal mine production, and larger load is added to the coal cost. At present, the variable frequency motors are mostly used in the coal industry, but the coal mine belts are more, the belts are dispersed, the purpose of energy conservation and efficiency improvement cannot be met by artificially adjusting the speed, meanwhile, gangue and anchor rods with different sizes can be randomly generated in the normal coal transportation process, and the transportation belts are damaged to cause unnecessary loss to customers if not cleaned.
In the prior art, a detection method based on a YOLO model is adopted to analyze a coal flow monitoring image for one time, the accuracy of the method for analyzing the abnormity of the coal flow monitoring image averagely reaches 93.7%, but the calculation efficiency of a graphic processor is low, and the calculation cost of the model is too high. The other prior art combines a fuzzy clustering algorithm and a Canny operator to realize the identification of the defects in the coal flow monitoring image, and the method can identify almost all defect edges, but also contains a large amount of redundant information, and brings certain noise interference to the coal flow monitoring image.
Disclosure of Invention
Aiming at the problems, the invention discloses an intelligent coal flow monitoring data analysis method for improving a CNN algorithm model, which can identify a coal flow monitoring image and realize the abnormity analysis and processing of coal flow monitoring.
In order to achieve the technical effects, the invention adopts the following technical scheme:
an intelligent coal flow monitoring data analysis method for improving a CNN algorithm model comprises the following steps:
(S1) collecting coal flow monitoring data by a spherical camera in a monitoring area by adopting a machine vision technology;
(S2) denoising and eliminating the collected coal flow monitoring image by an improved dark channel prior filtering method;
as a further technical scheme of the invention, the improved dark channel prior filtering method comprises the following steps:
(S21) decomposing and identifying the collected coal flow monitoring image, wherein the identification time of each section is recorded as:
in the formula (1), t (x) represents the identification time for decomposing the coal flow monitoring image, x represents the coal flow monitoring image, omega represents a fuzzy denoising constant, I represents the coal flow monitoring image noise, and A represents the influence factor of the coal flow monitoring image by the external environment;
(S22) processing the coal flow monitoring image in the identification time to obtain an identification result:
in the formula (2), J (x) represents the recognition result after the coal flow monitoring image processing, t 0 Representing the initial recognition duration of the coal flow monitoring image;
(S23) according to the coal flow monitoring image recognition result, carrying out tolerance calculation on the fuzzy region of the coal flow monitoring image:
δ(x)＝|I-A| (3)
in the formula (3), δ (x) represents a fuzzy region tolerance calculation amount of the coal flow monitoring image;
(S24) counting the fuzzy region tolerance of all the coal flow monitoring images, and then removing the fuzzy region tolerance in a dark channel mode, wherein the denoising and removing result of the coal flow monitoring images is output as follows:
in the formula (4), G (x) represents a coal flow monitoring image denoising result output by an improved dark channel prior method;
(S3) the coal flow monitoring image subjected to denoising processing is transmitted to a data storage center through a network in an indirect mode, and an improved CNN algorithm model is adopted to carry out data analysis on the coal flow monitoring image;
as a further technical scheme of the invention, the improved CNN algorithm model comprises the following steps:
(S31) firstly, a CNN algorithm model is constructed, a coal flow monitoring image x is input, and h is assumed 1 And h 2 Respectively the length and the width of the I-th layer convolution kernel of the CNN algorithm model, and the CNN algorithm model convolution layer output function constructed by j coal flow monitoring images:
in the formula (5), g represents a convolutional layer output function, f represents an activation function of a convolutional layer, and v represents the position of a jth coal flow monitoring image feature map of a first convolutional layer kernel; iteratively training the convolutional layer through a formula (5) to construct a CNN algorithm model, and setting dynamic adaptive parameters of the CNN algorithm model as follows:
w 0 ＝w 1 ·γ c +w 2 (1-γ c ) (6)
in formula (6), w 0 Representing the current weight value, w, of the CNN algorithm model 1 Initial weight balance value, w, representing CNN algorithm model 2 Representing the final weight of the CNN algorithm model, γ c Index parameters representing weight factors of the CNN algorithm model, and c represents the iterative training times of the CNN algorithm model;
(S32) improving the CNN algorithm model through the convolution encoder auxiliary model, outputting an optimized objective function by the convolution encoder auxiliary model, and turning the coal flow monitoring image with high and low mixed dimensionality to a low latitude coal flow monitoring image:
M(θ)＝∑L[P,f(P)]
(7)
in the formula (7), theta represents the dimension of a low-dimensional vector of the coal flow monitoring image, M represents a target function of the coal flow monitoring image dimension reduction conversion, and P represents a variable of the convolution encoder in the coal flow monitoring image dimension reduction conversion;
(S33) setting 7-12 hidden layer nodes of the improved CNN algorithm model through multi-scale feature fusion and a single-excitation multi-box detector algorithm, and setting a weight vector and a threshold vector according to an input coal flow monitoring image, wherein the abnormal probability q of the coal flow monitoring image is as follows:
in the formula (8), a represents a multi-scale feature fusion and single-excitation multi-box detector algorithm threshold parameter, and i represents a coal flow monitoring image type ordinal number; formula (8) represents the probability of the improved CNN algorithm model in all the images monitored by the coal flow, wherein the abnormal images are occupied;
(S34) calculating a fitness function f of the coal flow monitoring abnormal image as follows:
in the formula (9), n represents the number of the types of the abnormal images for monitoring the coal flow; outputting cumulative probability P of abnormal images for monitoring coal flow of the c-th iterative training c Comprises the following steps:
in the formula (10), k represents a constant parameter, f avg Representing the fitness function average value of the coal flow monitoring abnormal image; f. of c The coal flow monitoring abnormal image fitness function of the c-th iterative training is represented, and the coal flow monitoring can be directly output through the calculation of formulas (5) to (10)Cumulative probability of image anomaly P c When P is c When the value of the intelligent coal flow monitoring image is close to 1, judging that the coal flow monitoring image is in an abnormal state, and finishing the abnormal analysis result of the intelligent coal flow monitoring image;
and (S4) transmitting the abnormal analysis result to a monitoring alarm module, and displaying the abnormal analysis result through an OLED display.
As a further technical solution of the present invention, in the step (S1), the machine vision technology employs a laser scanning mode to emit onto a coal flow target to be collected at an angle with a dome camera, then rapidly determines the height of an illumination point through the deviation of a laser thin line in an image of the dome camera, scans the laser line pattern on the coal flow target to form a surface profile of the coal flow target, and completes the collection of a coal flow monitoring image.
As a further technical scheme of the invention, in the step (S2), the external environmental influence factors of the coal flow monitoring image comprise noise generated by a coal briquette pulverizer, noise generated by the contact among coal flow particles, noise generated by the random motion of optical particles of a laser technology and noise generated by a hardware circuit in the coal flow monitoring operation process.
As a further technical scheme of the invention, in the step (S3), the coal flow monitoring image with high and low mixed dimensionality comprises a three-dimensional coal flow monitoring image, a two-dimensional coal flow monitoring image and a one-dimensional coal flow monitoring image.
As a further technical solution of the present invention, an intelligent coal flow monitoring system includes:
the monitoring area I is used for monitoring the surrounding environment state of the coal flow transportation production line, the monitoring area 1 comprises an intelligent terminal, the intelligent terminal is provided with three sensors of MQ-135, MQ-9 and MQ-4, and the three sensors of MQ-135, MQ-9 and MQ-4 are used for collecting CO and CO of the coal flow transportation production line 2 、CH 4 The gas concentration is further measured, monitored, protected and controlled in a pressure stabilizing way; the monitoring area I uniformly schedules the whole monitoring area through a management center, transmits data to each workstation through a data I channel, and eliminates faults through equipment protection;
monitoring a second area; driving operation for monitoring coal flow transportation production line, adopted drivingThe moving mode is mechanical driving soft start, the starting time can be adjusted according to the main parameters of the belt conveyor, the conveyor is started stably, full load starting can be realized, and the soft driving controls the starting tension to enable the starting acceleration value of the conveyor belt to be 0.5m/s 2 (ii) a The monitoring area II carries out coal flow monitoring through a spherical camera of TYNJK-TL0661 specification, the spherical camera adopts a machine vision technology to collect coal flow monitoring images, the collected coal flow monitoring images are denoised and eliminated through an improved dark channel prior filtering method, and then the processed coal flow monitoring images are transmitted to a data storage center through a network;
a monitoring alarm module; the monitoring and alarming module is controlled by a CPU, can be easily programmed in an Arduino development environment running on a computer and can be uploaded to the monitoring and alarming module;
a data storage center; the data storage center is connected with the monitoring area I and the monitoring area II through a network for data interaction; the data storage center performs anomaly analysis on the coal flow monitoring image through an improved CNN algorithm model module;
the improved CNN algorithm model module is coupled to the data storage center, and the data storage center is connected to the monitoring alarm module.
As a further technical solution of the present invention, the monitoring and warning module includes:
the CPU is used for controlling the monitoring alarm module; the CPU is an ATmega328P type open source chip, reads and writes output coal flow monitoring image signals, and a circuit board of the CPU comprises a 16MHz crystal oscillator and fourteen digital input or output pins, wherein six pulse width modulation pins are included;
a detector for receiving communication signals from the direct-connect sensor and the remote slave sensor;
the alarm setting switch is used for starting or closing the working state of the monitoring alarm module;
the OLED display is used for displaying the abnormal analysis result of the coal flow monitoring image; the OLED display is in two-way communication with the CPU and comprises a programming interface related to the CPU, and the programming interface is connected with the CPU according to the USB3.0 specification;
the channel alarm relay is used for various data input devices, and the monitoring alarm module is activated by the 4-channel alarm relay; the sensor provides about 3 programmable alarm levels for each channel, an audible alarm with a mute function, a red LED flash lamp and an alarm setting switch;
the electrostatic discharge locking valve is used for closing the operation of the monitoring alarm module, can respond to a manual alarm setting switch to carry out closing alarm operation, determines that the dangerous conditions of the network coal flow data do not exist any more and then carries out manual reset, and resets the operation of the monitoring alarm module; the electrostatic discharge latching valve works with a four-channel alarm relay that receives an alarm signal from a sensor and is actuated by latching (closing) the solenoid valve with a 50 millisecond electrical pulse.
As a further technical scheme of the invention, the network access adopts a narrow-band Internet of things technology and is constructed into a cellular network; the network consumes about 180kHz bandwidth indirectly, and can be directly deployed in a GSM network.
As a further technical scheme of the invention, the monitoring area I is used for monitoring the ambient environment state of the coal flow transportation production line through coal, CO and CO 2 、CH 4 The gas mixtures are combined into a gas-coal mixture and the coal flow is determined by subtracting the gas flow from the mixture flow.
As a further technical solution of the present invention, the improved dark channel prior filtering method includes a noise detection module, where the noise detection module is a phase similarity detection method, and calculates similarities of different coal flow monitoring data information sets, which can be expressed as:
in the formula (11), E ti Data set representing different coal flow monitoring data of data to be measuredAnd E is ti Where ti is the identity of the actual measurement, E t Representing data sets of different coal flow monitoring data in the template, t representing the identifier of the template data information, lambda (e) ij ,G t ) Weight coefficient representing the edge after correction, a ij Expressing elements in the data information correlation matrix, and [ mu ] (i, j) expressing the correlation degree of the data information, calculating the correlation degree of the data to be detected and different coal flow monitoring data information of the template through a formula (11), and expressing the matching similarity between the template and the data to be detected together with the formula (11); and filtering the data information when the data are similar, and keeping the data information when the data are not similar.
The invention has the beneficial and positive effects that:
the method is different from the conventional technology, can quickly analyze the abnormal condition of the coal flow monitoring image and work at the minimum cost, and reduces the interference of noise on the coal flow monitoring image by denoising and eliminating the coal flow monitoring image acquired by the machine vision technology, so that the obtained coal flow monitoring image is clearer.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the prior art descriptions will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without inventive labor, wherein:
FIG. 1 shows a flow chart of an intelligent coal flow monitoring data analysis method of an improved CNN algorithm model;
FIG. 2 illustrates a block diagram of an intelligent coal flow monitoring system;
FIG. 3 shows an internal structure diagram of the monitoring alarm module;
FIG. 4 shows a training error comparison of the improved CNN algorithm model to the conventional CNN algorithm model;
FIG. 5 shows a comparison graph of the coal flow monitoring image anomaly identification precision by three methods;
Detailed Description
The preferred embodiments of the present invention will be described in conjunction with the accompanying drawings, it being understood that the embodiments described herein are merely illustrative and explanatory of the invention, and are not restrictive thereof;
as shown in fig. 1, an intelligent coal flow monitoring data analysis method for improving a Convolutional Neural Network (CNN) algorithm model includes the steps of:
(S1) collecting coal flow monitoring data by a ball-shaped camera in a monitoring area by adopting a machine vision technology;
in a specific embodiment, the machine vision industrial inspection system is divided into two main categories of quantitative and qualitative inspection, each of which is divided into different sub-categories, in terms of its inspection properties and application range. Machine vision is very active in various application fields of industrial online detection, such as: visual inspection of a printed circuit board, automatic flaw detection of the surface of a steel plate, measurement of parallelism and verticality of a large workpiece, detection of container volume or impurities, automatic identification and classification of mechanical parts, measurement of geometrical dimensions and the like. Furthermore, in many other applications where the method is difficult to detect, it can be effectively implemented using a machine vision system. The use of machine vision is increasingly replacing human beings to perform a number of tasks, which undoubtedly greatly increases the level of production automation and the level of intelligence of detection systems. Machine vision is different from computer vision, and it relates to image processing, artificial intelligence and pattern recognition machine vision is a project dedicated to the collection of mechanical, optical, electronic, software systems, inspection of natural objects and materials, artificial defects and manufacturing processes, and it is intended to detect defects and improve quality, operational efficiency, and product and process safety. It is also used to control machines.
(S2) denoising and eliminating the collected coal flow monitoring image by an improved dark channel prior filtering method;
in a specific embodiment, the improved dark channel prior filtering method comprises the following steps:
(S21) decomposing and identifying the collected coal flow monitoring image, wherein the identification time of each section is recorded as:
in the formula (1), t (x) represents the identification time for decomposing the coal flow monitoring image, x represents the coal flow monitoring image, omega represents a fuzzy denoising constant, I represents the coal flow monitoring image noise, and A represents the influence factor of the coal flow monitoring image by the external environment;
in a specific embodiment, in a coal flow monitoring image, the lower the fuzziness of the coal flow monitoring image is, the darker a dark channel image thereof is, and the smaller the pixel point value is; the higher the coal flow monitoring image fuzziness is, the brighter the dark channel image is, and the larger the pixel point value is, so that the dark channel image can better reflect image fuzzy information. The selection method for the atmospheric light value comprises the following steps: firstly, selecting the pixels with the maximum pixel value of the first 0.1% in the dark channel image, wherein the pixels correspond to the coal flow monitoring image, and selecting the coal flow monitoring image with the highest pixel value.
(S22) processing the coal flow monitoring image in the identification time to obtain an identification result:
in the formula (2), J (x) represents the recognition result after the coal flow monitoring image processing, t 0 Indicating the initial recognition duration of the coal flow monitoring image.
In a specific embodiment, the coal flow monitoring image processing is to sort out each character image and send the character image to the recognition module for recognition, and the process is called image preprocessing. In image analysis, the input image is subjected to processing performed before feature extraction, segmentation, and matching. The main purposes of image preprocessing are to eliminate irrelevant information from the image, recover useful real information, enhance the detectability of relevant information and simplify the data to the maximum extent, thereby improving the reliability of feature extraction, image segmentation, matching and recognition.
(S23) according to the coal flow monitoring image recognition result, carrying out tolerance calculation on the fuzzy region of the coal flow monitoring image:
δ(x)＝|I-A| (3)
in equation (3), δ (x) represents a blur area tolerance calculation amount of the coal flow monitor image.
In one embodiment, the image noise is not spatial in nature, i.e., the point is not abrupt with respect to surrounding points, i.e., the point is noisy. But the point is relative to the different points generated at the same position in continuous time, if the error is large, the point can be called as noise, namely the noise is time domain in nature. Then, when we calculate a certain region of an image, we sometimes use a flat image of the certain region to calculate the snr of the certain region, and actually use a potential assumption: the points of the flat area can be considered as a collection in a continuous time of the center point.
(S24) counting the fuzzy region tolerance of all the coal flow monitoring images, and then removing the fuzzy region tolerance in a dark channel mode, wherein the denoising and removing result of the coal flow monitoring images is output as follows:
in the formula (4), G (x) represents a coal flow monitoring image denoising result output by an improved dark channel prior method;
in a specific embodiment, noise is an important cause of image interference, and an image may have various noises in practical application, and these noises may be generated in transmission, quantization and the like. The noise and the signal can be divided into three forms of an original image (B), an image signal (C) and noise (D) according to the relation between the noise and the signal: (1) Additive noise, which is irrelevant to the input image signal, the noisy image can be expressed as B = C + D, and the noise generated when the camera of the light guide camera tube scans the image belongs to the noise; (2) Multiplicative noise, which is related to the image signal, noisy images can be represented as B = C + CD, noise when the flying spot scanner scans the image, coherent noise in the television image, and grain noise in the film. (3) Quantization noise, which is irrelevant to the input image signal, is generated by the quantization error existing in the quantization process and then reflected to the receiving end.
(S3) the coal flow monitoring image subjected to denoising processing is transmitted to a data storage center through a network in an indirect mode, and an improved CNN algorithm model is adopted to carry out data analysis on the coal flow monitoring image;
in a specific embodiment, the improved CNN algorithm model step includes:
(S31) firstly, a CNN algorithm model is constructed, a coal flow monitoring image x is input, and h is assumed 1 And h 2 The length and the width of the I layer convolution kernel of the CNN algorithm model are respectively, and the CNN algorithm model convolution layer output function constructed by j coal flow monitoring images together:
in the formula (5), g represents a convolutional layer output function, f represents an activation function of a convolutional layer, and v represents the position of the jth coal flow monitoring image feature map of the ith convolutional kernel;
in one embodiment, the convolutional layer functions to perform feature extraction on input data, and includes a plurality of convolutional kernels, where each element constituting a convolutional kernel corresponds to a weight coefficient and a deviation amount, and is similar to a neuron of a feedforward neural network. Each neuron in the convolution layer is connected to a plurality of neurons in a closely located region in the previous layer, the size of which depends on the size of the convolution kernel, known in the literature as the "receptive field", which has a meaning analogous to that of the visual cortical cells. When the convolution kernel works, the convolution kernel regularly sweeps the input characteristics, matrix element multiplication summation is carried out on the input characteristics in the receptive field, and deviation amount is superposed. Convolutional layers composed of unit convolutional kernels are also referred to as net-in-net or multi-layer perceptron convolutional layers. The unit convolution kernel can reduce the number of channels of the graph while maintaining the feature size, thereby reducing the amount of computation of the convolution layer. The convolutional neural network completely constructed by the unit convolutional kernel is a multilayer perceptron comprising parameter sharing.
In a specific embodiment, the convolutional layer parameters include convolutional kernel size, step size and padding, which together determine the size of the convolutional layer output feature map, and are hyper-parameters of the convolutional neural network. Where the convolution kernel size can be specified as an arbitrary value smaller than the input image size, the larger the convolution kernel, the more complex the input features that can be extracted. The convolution step defines the distance between the positions of two adjacent scanned feature maps of the convolution kernel, when the convolution step is 1, the convolution kernel scans the elements of the feature maps one by one, and when the step is n, n-1 pixels are skipped in the next scanning. As can be seen from the cross-correlation calculation of the convolution kernels, the size of the feature map gradually decreases with the stacking of convolution layers, for example, a 16 × 16 input image outputs a 12 × 12 feature map after passing through a unit step size, unfilled 5 × 5 convolution kernel. To this end, padding is a method of artificially increasing the size of the feature map before it passes through the convolution kernel to offset the effects of size shrinkage in the computation. The usual filling methods are filling at 0 and repeated boundary values.
In a specific embodiment, a convolutional layer is iteratively trained by a formula (5) to construct a CNN algorithm model, and dynamic adaptive parameters of the CNN algorithm model are set as follows:
w 0 ＝w 1 ·γ c +w 2 (1-γ c ) (6)
in the formula (6), w 0 Representing the current weight value, w, of the CNN algorithm model 1 Initial weight balance value, w, representing CNN algorithm model 2 Representing the final weight of the CNN algorithm model, γ c Index parameters representing weight factors of the CNN algorithm model, and c represents the iterative training times of the CNN algorithm model;
in a specific embodiment, the CNN algorithm model is a type of feedforward neural network including convolution calculation and having a deep structure, and is one of the representative algorithms of deep learning. The CNN algorithm model has the characteristic learning capacity and can carry out translation invariant classification on input information according to the hierarchical structure of the CNN algorithm model, so the CNN algorithm model is also called as a translation invariant artificial neural network CNN algorithm model to imitate the visual perception mechanism construction of organisms and can carry out supervised learning and unsupervised learning, and the convolution kernel parameter sharing in the hidden layer and the sparsity of interlayer connection enable the convolutional neural network to carry out learning on lattice characteristics such as pixels and audio with smaller calculated amount, thereby having stable effect and having no additional characteristic engineering requirements on data.
In the specific embodiment, in order to enable the accuracy of the class characteristics of the coal flow monitoring image to be higher, image training data sets such as an MNIST (public network integration test), an ImageNet data set, a flame data set, a PASCAL VOC (continuous operational simulation) training set and a COCO (chip on co) are adopted, and a visual computing method is applied to realize sample multiplication and expansion of the size, rotation, angle, RGB (red, green and blue) effect and GAN (global anti-neural network) algorithm of the coal flow monitoring image, so that a training model meets the requirement of large-data-volume training, and therefore each large class characteristic is obtained. The core of the method is that the feature extraction and the classifier are organically integrated, back propagation is carried out in a random gradient descent mode, and the parameters of the convolution template and the parameters of the full connection layer are continuously optimized, so that the finally learned features and the classifier are close to the optimum, and the object features are obtained.
(S32) improving the CNN algorithm model through the convolution encoder auxiliary model, outputting an optimized objective function by the convolution encoder auxiliary model, and turning the coal flow monitoring image with high and low mixed dimensionality to a low latitude coal flow monitoring image:
M(θ)＝∑L[P，f(P)]
(7)
in the formula (7), theta represents the dimension of a low-dimensional vector of the coal flow monitoring image, M represents a target function of the coal flow monitoring image dimension reduction conversion, and P represents a variable of the convolution encoder in the coal flow monitoring image dimension reduction conversion;
in the specific embodiment, the feature recognition and classification mainly solves the problem of independence of input information of 25 frames/s of time sequence monitoring video images, and realizes a CNN algorithm model with memory capacity required for continuous tracking of video feature target objects. The CNN algorithm model realizes continuous identification of features of a target object in front and back frame images, optimizes and reduces intermediate iterative regression convolution processes in order to avoid low space utilization rate, and performs softmax classification and position regression simultaneously when detecting on a plurality of feature maps. The prediction of the position areas and the class probabilities of a plurality of target objects allows the classification layer to reuse the characteristic graph characteristics generated at a plurality of image resolutions, forms high-speed continuous recursive image target identification and improves the identification accuracy. And improving the image segmentation effect by using a low layer, and performing region prediction by adopting feature mapping of a proper added lower layer. The granularity of image target feature vector detection is improved, the image segmentation and object detection performance is improved, and better identification quality and accuracy are obtained. Since the positioning error of the loss function is the main reason for influencing the detection effect, the position positioning accuracy value and the category score confidence coefficient loss function are fused together. The method comprises the steps of preprocessing an image training library by using a model, carrying out matching detection on an object in an image by combining a priori specific target characteristic vector obtained from the image under a line, and realizing the specific target detection and tracking functions of the image by circularly updating. And finally, real-time monitoring based on the coal flow is realized.
(S33) setting 7-12 hidden layer nodes of the improved CNN algorithm model through multi-scale feature fusion and a single-excitation multi-box detector algorithm, and setting a weight vector and a threshold vector according to an input coal flow monitoring image, wherein the abnormal probability q of the coal flow monitoring image is as follows:
in the formula (8), a represents a multi-scale feature fusion and single-excitation multi-box detector algorithm threshold parameter, and i represents a coal flow monitoring image type ordinal number; formula (8) represents the probability of the improved CNN algorithm model in all the images monitored by the coal flow, wherein the abnormal images are occupied;
in particular embodiments, to improve the CNN algorithm model, the prior art includes phase Fast-RCNN, and the like. These algorithms provide more accurate results for target detection. They appear somewhat slow for real-time detection. SSDs come at this time with a good balance between accuracy and computation speed. The structure of the SSD is based on VGG-16. But here some minor adjustments to VGG-16 are made, starting from the Conv6 layer, we replace the original fully connected layer with a series of auxiliary convolutional layers. Because the VGG-16 can provide high quality image classification and transfer learning to improve results, we take it as the underlying network for SSD. By using the auxiliary convolutional layers, we can extract features at multiple scales of the image and gradually reduce the size of each convolutional layer. I have discussed its working principle in the next section. You can see the following image of the VGG-16 architecture, which contains the full connectivity layer.
(S34) calculating a fitness function f of the coal flow monitoring abnormal image as follows:
in the formula (9), n represents the number of the types of the abnormal images for monitoring the coal flow; outputting cumulative probability P of abnormal images for monitoring coal flow of the c-th iterative training c Comprises the following steps:
in the formula (10), k represents a constant parameter, f avg Representing the fitness function average value of the coal flow monitoring abnormal image; f. of c The fitness function of the coal flow monitoring abnormal image of the c-th iterative training is shown, and the abnormal cumulative probability P of the coal flow monitoring image can be directly output through the calculation of formulas (5) to (10) c When P is c And when the value of the intelligent coal flow monitoring image is close to 1, judging that the coal flow monitoring image is in an abnormal state, and finishing the abnormal analysis result of the intelligent coal flow monitoring image.
In a particular embodiment, deep learning techniques are applied to detect anomalies in images, the idea behind which is to bypass an automatic encoder neural network in which the encoder takes input data and compresses it into a latent space representation. The decoder will then reconstruct the input data from this space. When training the hidden layer, it has been taught how to denoise the input data, which is the key to detect anomalies. When we apply this network to anomalous images, the reconstructed image will have a higher mean square error since the data samples have never been seen before by the auto-encoder. While this is a new effective method of detecting anomalies, it has some negative impact, e.g., we should choose the MSE threshold and should train on a particular class of data on the network, while anomalies should be of a completely different class and therefore can be detected as anomalies.
And (S4) transmitting the abnormal analysis result to a monitoring alarm module, and displaying the abnormal analysis result through an OLED display.
In a specific embodiment, a machine vision technology adopts a laser scanning mode to emit the laser scanning mode to a dome camera at a certain angle to a coal flow target to be collected, then the height of an illumination point is rapidly determined through the deviation of a laser thin line in an image of the dome camera, the laser line pattern is scanned on the coal flow target to form the surface outline of the coal flow target, and the collection of a coal flow monitoring image is completed. In addition, the laser imaging radar adopts a Q-switched pulse DPL laser as a transmitting source, laser pulses are transmitted through scanning, a unit receiving device receives echo signals through scanning, and after signal preprocessing, an image acquisition processing system stores, processes and displays image data according to the synchronous sequence of fields, lines and columns. The whole imaging process is continuous, which requires that the image acquisition/processing system must continuously acquire, process and display each frame of image, i.e. achieve real-time acquisition, processing and display of frames.
In an embodiment, the external environmental influence factors of the coal flow monitoring image include noise generated by a coal pulverizer, noise generated by contact between coal flow particles, noise generated by the randomness of optical particle motion of a laser technology, and noise generated by hardware circuits in a coal flow monitoring operation process. In addition, the coal flow monitoring image with high and low mixed dimensions comprises a three-dimensional coal flow monitoring image, a two-dimensional coal flow monitoring image and a one-dimensional coal flow monitoring image.
In a specific embodiment, the intelligent coal flow monitoring system comprises a monitoring area I, a monitoring alarm module and a data storage center. The monitoring area I is used for monitoring the surrounding environment state of the coal flow transportation production line, the monitoring area 1 comprises an intelligent terminal, the intelligent terminal is provided with three sensors of MQ-135, MQ-9 and MQ-4, and the three sensors of MQ-135, MQ-9 and MQ-4 are used for collecting CO and CO of the coal flow transportation production line 2 、CH 4 The gas concentration is further measured, monitored, protected and controlled in a pressure stabilizing way; the monitoring I area is used for monitoring the whole system through the management centerAnd uniformly scheduling the monitoring areas, transmitting data to each workstation through a data I channel, and removing faults through equipment protection. The monitoring area II is used for monitoring the driving operation of a coal flow transportation production line, the adopted driving mode is mechanical driving soft start, the starting time can be adjusted according to the main parameters of the belt conveyor, the conveyor is stably started, full load starting can be realized, and the soft driving controls the starting tension to enable the starting acceleration value of the conveyor belt to be 0.5m/s 2 (ii) a And the monitoring II area carries out coal flow monitoring through a spherical camera of TYNJK-TL0661 specification, the spherical camera adopts a machine vision technology to collect a coal flow monitoring image, the collected coal flow monitoring image is subjected to denoising elimination through an improved dark channel prior filtering method, and then the processed coal flow monitoring image is transmitted to a data storage center through network interconnection.
In a specific embodiment, the monitoring alarm module is used for performing sound-light alarm on abnormal phenomena in the coal flow monitoring image, is controlled by the CPU, can be easily programmed in an Arduino development environment running on a computer, and can be uploaded to the monitoring alarm module; the data storage center is used for storing the coal flow environment data acquired by the monitoring area I and the coal flow monitoring image acquired by the monitoring area II, and the data storage center is connected with the monitoring area I and the monitoring area II through a network in an interconnecting way for data interaction; and the data storage center performs anomaly analysis on the coal flow monitoring image through an improved CNN algorithm model module. In addition, the monitoring I area and the monitoring II area are connected with a data storage center through network interconnection, an improved CNN algorithm model module is coupled on the data storage center, the data storage center is connected with a monitoring alarm module, and as the management requirement has low requirements on the identification of moving objects and the accuracy of target characteristics, in order to ensure the real-time retrieval of moving target objects, PYTHON + openCV technology is applied to carry out on-line preprocessing on coal flow monitoring images, so that the frame reduction and resolution reduction processing are completed, the calculated amount is reduced, and the calculation speed is improved. The method is characterized in that the continuous retrieval of target features is realized by applying an improved CNN algorithm model, and the moving target identification and tracking of relevant features such as position, speed, shape and the like are established among continuous image frames.
In a specific embodiment, the monitoring alarm module includes: the CPU is used for controlling the monitoring alarm module; the CPU is an ATmega328P type open source chip, reads and writes output coal flow monitoring image signals, and a circuit board of the CPU comprises a 16MHz crystal oscillator and fourteen digital input or output pins, wherein six pulse width modulation pins are included; a detector for receiving communication signals from the direct-connect sensor and the remote slave sensor; the alarm setting switch is used for starting or closing the working state of the monitoring alarm module; the OLED display is used for displaying the abnormal analysis result of the coal flow monitoring image; the Organic Light-Emitting Diode (OLED) display is in two-way communication with the CPU and comprises a programming interface related to the CPU, and the programming interface is connected with the CPU according to the USB3.0 specification; the organic light emitting display belongs to a new rising type on the LCD of the mobile phone and is known as a 'dream display'. OLEDs are also known as third generation display technologies. The OLED is thinner, low in energy consumption, high in brightness, good in luminous efficiency, capable of displaying pure black, and capable of being bent, such as current curved screen televisions, mobile phones and the like. Nowadays, international manufacturers strive for terribly and strongly invest in the research and development of the OLED technology, so that the OLED technology is more and more widely applied in the fields of television, computer, mobile phone, tablet and the like.
In a specific embodiment, the four-channel alarm relay is used for a plurality of data input devices, and the monitoring alarm module is activated by the 4-channel alarm relay; the sensor provides about 3 programmable alarm levels for each channel, an audible alarm with a mute function, a red LED flash lamp and an alarm setting switch; the electrostatic discharge locking valve is used for closing the operation of the monitoring alarm module, can respond to a manual alarm setting switch to perform closing alarm operation, and is manually reset after determining that the dangerous conditions of the network coal flow data do not exist any more, so that the monitoring alarm module is reset to operate; the electrostatic discharge latching valve works with a four-channel alarm relay that receives an alarm signal from a sensor and is actuated by latching (closing) the solenoid valve with a 50 millisecond electrical pulse.
In a specific embodiment, the network access adopts a narrowband internet of things technology to construct a cellular network; the network consuming a medium of about 180kHzBandwidth, can be directly deployed in a Global System for Mobile communications (GSM) network. The monitoring I area is used for monitoring the ambient environment state of the coal flow transportation production line, and the coal, CO and CO are used 2 、CH 4 The gas mixture is combined into a gas-coal mixture and the coal flow rate is determined by subtracting the gas flow rate from the mixture flow rate.
In a particular embodiment, the goal of GSM security is to make the system as secure as the public switched telephone network. The radio path system in the system is the weakest part because the radio signal can be easily intercepted. Mobile stations have a well known security problem: eavesdropping using the MS is technically possible (e.g., as an "eavesdropper"). Even if it has been switched off, it can be switched on via the air interface, so the best protection method is to take the battery out. The GSM MoU group considers that the technical characteristics of security are only a small part of the security requirements, and the greatest threat comes from simpler attacks such as leakage of encryption keys and insecure billing systems. Therefore, there are complex aspects to be taken to ensure that these security procedures meet the security requirements, and in addition, the cost-effectiveness of the security measures must be considered. The security requirements of the GSM system take into account some potential weaknesses of the cellular network. The security of the system should be appropriate for both the system operator and the user. The system operator wants to be able to ensure that the correct person is charged and that the service is not affected; the user requires privacy to be protected. For this purpose the GSM system of the operator must be designed taking into account environmental and security procedures, such as the generation and distribution of keys, the exchange of information between operators and the confidentiality of algorithms.
In a specific embodiment, in order to verify the performance of the algorithm provided by the invention, an improved CNN algorithm model adopted by the invention is subjected to simulation comparison, an abnormal evaluation result in the intelligent coal flow monitoring process is analyzed in an investigation mode to complete recording, the abnormal evaluation result is subjected to system overall model evaluation, training errors of the traditional CNN algorithm model are compared, and after simple data preprocessing, the training errors of two methods of iteration for 100 times are counted, so that the comparison of convergence is obtained as shown in fig. 3.
In the above embodiment, the improved dark channel prior filtering method includes a noise detection module, where the noise detection module is a phase similarity detection method, and calculates similarities of different coal flow monitoring data information sets, which may be represented as:
in the formula (11), E ti Data sets representing different coal flow monitoring data of the data to be measured, E ti Where ti is the identity of the actual measurement, E t Representing data sets of different coal flow monitoring data in the template, t representing the identifier of the template data information, lambda (e) ij ,G t ) Weight coefficient representing edge after correction, a ij Expressing elements in the data information correlation matrix, and mu (i, j) expressing the correlation degree of the data information, calculating the correlation degree of the data to be detected and the different coal flow monitoring data information of the template through a formula (11), and expressing the matching similarity between the template and the data to be detected together with the formula (11); filtering the data information when the data are similar, and keeping the data information when the data are dissimilar; information filtering is performed by the above method.
In the specific embodiment, in order to verify the performance of the algorithm provided by the invention, the improved CNN algorithm model adopted by the invention is subjected to simulation comparison, an abnormal judgment result in the intelligent coal flow monitoring process is analyzed in an investigation mode to complete recording, the image recognition precision operation of the whole system model is carried out on the abnormal judgment result, and the result is expressed in the form of a simulation curve, so that the simulation and the analysis are completed. The hardware configuration CPU of the computer used in the simulation process is Intercore i7-9700H, the operation memory is 3200MHz 8 multiplied by 2GB, and the size of the hard disk is 1TB. The simulation comparison is carried out by adopting the method of the invention to compare with a scheme I (a detection method based on a YOLO model) and a scheme II (a fuzzy clustering algorithm and a Canny operator fusion algorithm), a dome camera is adopted to collect 1024MB coal flow monitoring images, the effectiveness of the research is verified according to the data result calculated by a microcomputer, the experimental result is summarized into a data table, and the final experimental data result of the abnormal analysis precision of the 256-1024 MB coal flow monitoring images is shown in the table 1.
TABLE 1 results of the experiment
In a specific embodiment, a comparison experiment is further completed by comparing the coal flow monitoring image abnormality identification precision of each method, simulation of the comparison process is realized according to MTALAB software, a coal flow monitoring image abnormality identification precision curve is obtained, for example, as shown in FIG. 4, comparison shows that the identification precision of three coal flow monitoring image abnormality analysis methods has certain fluctuation, but the improved CNN algorithm model adopted by the method has the highest identification precision of 97.536%, which is far higher than that of the other two scheme methods, the requirement on coal flow monitoring image abnormality processing is met, and the coal flow monitoring image abnormality identification precision is improved.
Although specific embodiments of the present invention have been described above, it will be understood by those skilled in the art that these specific embodiments are merely illustrative and that various omissions, substitutions and changes in the form and details of the methods and systems described above may be made by those skilled in the art without departing from the spirit and scope of the invention; for example, it is within the scope of the present invention to combine the steps of the methods described above to perform substantially the same function in substantially the same way to achieve substantially the same result; accordingly, the scope of the invention is to be limited only by the following claims.
Claims (10)
1. An intelligent coal flow monitoring data analysis method for improving a CNN algorithm model comprises the following steps:
(S1) collecting coal flow monitoring data by a ball-shaped camera in a monitoring area by adopting a machine vision technology;
(S2) the collected coal flow monitoring image is filtered a priori through an improved dark channelDenoising and eliminating by the method; the improved dark channel prior filtering method comprises a noise detection module, wherein the noise detection module is a phase similarity detection method and is used for calculating the similarity of different coal flow monitoring data information sets, and a similarity function is expressed as follows:
wherein E ti Data sets representing different coal flow monitoring data of the data to be measured, E ti Where ti is the identity of the actual measurement, E t Data sets representing different coal flow monitoring data in the template, t represents the identifier of template data information, and lambda (e) ij ,G t ) Weight coefficient representing the edge after correction, a ij Representing elements in the data information correlation matrix, and mu (i, j) representing the correlation degree of the data information, calculating the correlation degree of the data to be detected and different coal flow monitoring data information of the template through a similarity function, and representing the matching similarity between the template and the data to be detected together with the similarity function; when the data are similar, filtering the data information, and when the data are dissimilar, retaining the data information;
(S3) the coal flow monitoring image subjected to denoising processing is transmitted to a data storage center through a network in an indirect mode, and an improved CNN algorithm model is adopted to carry out data analysis on the coal flow monitoring image;
and (S4) transmitting the abnormal analysis result to a monitoring alarm module, and displaying the abnormal analysis result through an OLED display.
2. The intelligent coal flow monitoring data analysis method for improving the CNN algorithm model as claimed in claim 1, wherein: in the step (S1), the machine vision technology adopts a laser scanning mode to emit the laser scanning mode to a dome camera at a certain angle to a coal flow target to be collected, then the height of an illumination point is rapidly determined through the deviation of a laser thin line in an image of the dome camera, the laser line pattern is scanned on the coal flow target to form the surface outline of the coal flow target, and the collection of a coal flow monitoring image is completed.
3. The intelligent coal flow monitoring data analysis method for improving the CNN algorithm model as claimed in claim 1, wherein: in the step (S2), the external environmental influence factors of the coal flow monitoring image comprise noise generated by a coal briquette pulverizer, noise generated by the contact among coal flow particles, noise generated by the randomness of the movement of optical particles of a laser technology and noise generated by a hardware circuit in the coal flow monitoring operation process.
4. The intelligent coal flow monitoring data analysis method for improving the CNN algorithm model as claimed in claim 1, wherein: the improved dark channel prior filtering method comprises the following steps:
(S21) decomposing and identifying the collected coal flow monitoring image, wherein the identification time of each section is recorded as:
in the formula (1), t (x) represents the identification time for decomposing the coal flow monitoring image, x represents the coal flow monitoring image, omega represents a fuzzy denoising constant, I represents the coal flow monitoring image noise, and A represents the influence factor of the coal flow monitoring image by the external environment;
(S22) processing the coal flow monitoring image in the identification time to obtain an identification result:
in the formula (2), J (x) represents the recognition result after the coal flow monitoring image processing, t 0 Representing the initial recognition duration of the coal flow monitoring image;
(S23) carrying out tolerance calculation on the fuzzy region of the coal flow monitoring image according to the coal flow monitoring image identification result:
∑(x)＝|I-A|(3)
in the formula (3), δ (x) represents the tolerance calculated amount of the fuzzy region of the coal flow monitoring image;
(S24) counting the fuzzy region tolerance of all the coal flow monitoring images, and then removing the fuzzy region tolerance in a dark channel mode, wherein the denoising and removing result of the coal flow monitoring images is output as follows:
in the formula (4), G (x) represents a denoising result of the coal flow monitoring image output by the improved dark channel prior method.
5. The intelligent coal flow monitoring data analysis method for improving the CNN algorithm model as claimed in claim 1, wherein: and (S3) in the step, the coal flow monitoring images with high and low mixed dimensions comprise three-dimensional coal flow monitoring images, two-dimensional coal flow monitoring images and one-dimensional coal flow monitoring images.
6. The intelligent coal flow monitoring data analysis method for improving the CNN algorithm model as claimed in claim 1, wherein: wherein the improved CNN algorithm model comprises the following steps:
(S31) firstly, a CNN algorithm model is built, a coal flow monitoring image x is input, and h is assumed 1 And h 2 Respectively the length and the width of the I-th layer convolution kernel of the CNN algorithm model, and the CNN algorithm model convolution layer output function constructed by j coal flow monitoring images:
in the formula (5), g represents a convolutional layer output function, f represents an activation function of a convolutional layer, and v represents the position of a jth coal flow monitoring image feature map of a first convolutional layer kernel; iteratively training the convolutional layer through a formula (5) to construct a CNN algorithm model, and setting dynamic adaptive parameters of the CNN algorithm model as follows:
w 0 ＝w 1 ·γ c +w 2 (1-γ c ) (6)
in the formula (6), w 0 Represents the current weight value, w, of the CNN algorithm model 1 Representing CNN algorithmic modelsInitial weight balance value of, w 2 Representing the final weight of the CNN algorithm model, γ c Index parameters representing weight factors of the CNN algorithm model, and c represents the iterative training times of the CNN algorithm model;
(S32) improving the CNN algorithm model through the convolution encoder auxiliary model, outputting an optimized objective function by the convolution encoder auxiliary model, and turning the coal flow monitoring image with high and low mixed dimensionality to a low latitude coal flow monitoring image:
M(θ)＝∑[P，f(P)] (7)
in the formula (7), theta represents the dimension of a low-dimensional vector of the coal flow monitoring image, M represents an objective function of the coal flow monitoring image dimension reduction conversion, and P represents a variable of the convolution encoder in the coal flow monitoring image dimension reduction conversion;
(S33) setting 7-12 hidden layer nodes of the improved CNN algorithm model through multi-scale feature fusion and a single-excitation multi-box detector algorithm, and setting a weight vector and a threshold vector according to an input coal flow monitoring image, wherein the abnormal probability q of the coal flow monitoring image is as follows:
in the formula (8), a represents a multi-scale feature fusion and single-excitation multi-box detector algorithm threshold parameter, and i represents a coal flow monitoring image type ordinal number; formula (8) represents the probability of the improved CNN algorithm model in all the images monitored by the coal flow, wherein the abnormal images are occupied;
(S34) calculating a fitness function f of the coal flow monitoring abnormal image as follows:
in the formula (9), n represents the number of the types of the abnormal images for monitoring the coal flow; outputting cumulative probability P of abnormal images for monitoring coal flow of the c-th iterative training c Comprises the following steps:
in the formula (10), k represents a constant parameter, f avg Representing the fitness function average value of the coal flow monitoring abnormal image; f. of c The fitness function of the coal flow monitoring abnormal image of the c-th iterative training is shown, and the abnormal cumulative probability P of the coal flow monitoring image can be directly output through the calculation of formulas (5) to (10) c When P is c And when the value of the intelligent coal flow monitoring image is close to 1, judging that the coal flow monitoring image is in an abnormal state, and finishing the abnormal analysis result of the intelligent coal flow monitoring image.
7. An intelligent coal flow monitoring system applying the intelligent coal flow monitoring data analysis method of the improved CNN algorithm model of any one of claims 1-6, characterized in that: the intelligent coal flow monitoring system comprises:
the monitoring area I is used for monitoring the surrounding environment state of the coal flow transportation production line, the monitoring area 1 comprises an intelligent terminal, the intelligent terminal is provided with three sensors of MQ-135, MQ-9 and MQ-4, and the three sensors of MQ-135, MQ-9 and MQ-4 are used for collecting CO and CO of the coal flow transportation production line 2 、CH 4 The gas concentration is further measured, monitored, protected and controlled in a pressure stabilizing way; the monitoring I area uniformly schedules the whole monitoring area through a management center, transmits data to each workstation through a data I channel, and eliminates faults through equipment protection;
and a monitoring area II for monitoring the driving operation of the coal flow transportation production line, wherein the adopted driving mode is mechanical driving soft start, the starting time can be adjusted according to the main parameters of the belt conveyor, so that the conveyor is stably started, full load starting can be realized, and the soft driving controls the starting tension to enable the starting acceleration value of the conveyor belt to be 0.5m/s 2 (ii) a The monitoring area II carries out coal flow monitoring through a spherical camera of TYNJK-TL0661 specification, the spherical camera adopts a machine vision technology to collect coal flow monitoring images, the collected coal flow monitoring images are denoised and eliminated through an improved dark channel prior filtering method, and then the processed coal flow monitoring images are transmitted to a data storage center through a network;
the monitoring and alarming module is used for performing sound-light alarm on abnormal phenomena in the coal flow monitoring image, is controlled by the CPU, can be easily programmed in an Arduino development environment running on a computer and can be uploaded to the monitoring and alarming module;
the data storage center is used for storing the coal flow environment data acquired by the monitoring area I and the coal flow monitoring image acquired by the monitoring area II, and the data storage center is connected with the monitoring area I and the monitoring area II through a network in an interconnecting way for data interaction; the data storage center performs anomaly analysis on the coal flow monitoring image through an improved CNN algorithm model module;
the improved CNN algorithm model module is coupled to the data storage center, and the data storage center is connected to the monitoring alarm module.
8. The intelligent coal flow monitoring system of claim 7, wherein: the monitoring alarm module comprises:
the CPU is used for controlling the monitoring alarm module; the CPU is an ATmega328P type open source chip, reads and writes output coal flow monitoring image signals, and a CPU circuit board comprises a 16MHz crystal oscillator and fourteen digital input or output pins, wherein the fourteen digital input or output pins comprise six pulse width modulation pins;
a detector for receiving communication signals from the direct-connect sensor and the remote slave sensor;
the alarm setting switch is used for starting or closing the working state of the monitoring alarm module;
the OLED display is used for displaying the abnormal analysis result of the coal flow monitoring image; the OLED display is in two-way communication with the CPU, the OLED display comprises a programming interface related to the CPU, and the programming interface is connected with the CPU according to the USB3.0 specification;
the channel alarm relay is used for various data input devices, and the monitoring alarm module is activated by the 4-channel alarm relay; the sensor provides about 3 programmable alarm levels for each channel, an audible alarm with a mute function, a red LED flash lamp and an alarm setting switch;
the electrostatic discharge locking valve is used for closing the operation of the monitoring alarm module, can respond to a manual alarm setting switch to carry out closing alarm operation, determines that the dangerous conditions of the network coal flow data do not exist any more and then carries out manual reset, and resets the operation of the monitoring alarm module; the electrostatic discharge latching valve works with a four-channel alarm relay that receives an alarm signal from a sensor and is actuated by latching the solenoid valve with a 50 millisecond electrical pulse.
9. The intelligent coal flow monitoring system of claim 7, wherein: the network access adopts a narrow-band Internet of things technology and is constructed into a cellular network; the network is directly deployed in a GSM network, consuming about 180kHz bandwidth.
10. The intelligent coal flow monitoring system of claim 7, wherein: the monitoring area I is used for monitoring the ambient environment state of the coal flow transportation production line, and coal, CO and CO are used 2 、CH 4 The gas mixture is combined into a gas-coal mixture and the coal flow rate is determined by subtracting the gas flow rate from the mixture flow rate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211605959.9A CN115953731A (en) | 2022-12-14 | 2022-12-14 | Intelligent coal flow monitoring data analysis method for improving CNN algorithm model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211605959.9A CN115953731A (en) | 2022-12-14 | 2022-12-14 | Intelligent coal flow monitoring data analysis method for improving CNN algorithm model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115953731A true CN115953731A (en) | 2023-04-11 |
Family
ID=87288804
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211605959.9A Pending CN115953731A (en) | 2022-12-14 | 2022-12-14 | Intelligent coal flow monitoring data analysis method for improving CNN algorithm model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115953731A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117033143A (en) * | 2023-10-08 | 2023-11-10 | 常州瑞阳液压成套设备有限公司 | Intelligent monitoring data transmission system and method based on running state of big data |
-
2022
- 2022-12-14 CN CN202211605959.9A patent/CN115953731A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117033143A (en) * | 2023-10-08 | 2023-11-10 | 常州瑞阳液压成套设备有限公司 | Intelligent monitoring data transmission system and method based on running state of big data |
| CN117033143B (en) * | 2023-10-08 | 2024-01-26 | 常州瑞阳液压成套设备有限公司 | Intelligent monitoring data transmission system and method based on running state of big data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kim et al. | Deep-hurricane-tracker: Tracking and forecasting extreme climate events | |
| CN106875373B (en) | Mobile phone screen MURA defect detection method based on convolutional neural network pruning algorithm | |
| CN111553929A (en) | Mobile phone screen defect segmentation method, device and equipment based on converged network | |
| US11803953B2 (en) | Manufacturing quality improvement through statistical root cause analysis using convolution neural networks | |
| Gan et al. | Online rail surface inspection utilizing spatial consistency and continuity | |
| CN107423747B (en) | A kind of conspicuousness object detection method based on depth convolutional network | |
| CN113838034B (en) | Quick detection method for surface defects of candy package based on machine vision | |
| CN110751087B (en) | EOF-based unmanned aerial vehicle signal identification system and method | |
| Tayeh et al. | Distance-based anomaly detection for industrial surfaces using triplet networks | |
| CN112001922B (en) | Method and apparatus for diagnosing defect of charged equipment | |
| CN109899281A (en) | A kind of fracturing unit Weak fault diagnostic method and device based on infrared thermal imagery | |
| CN114241370A (en) | Intrusion identification method and device based on digital twin transformer substation and computer equipment | |
| CN115953731A (en) | Intelligent coal flow monitoring data analysis method for improving CNN algorithm model | |
| CN115131325A (en) | Breaker fault operation and maintenance monitoring method and system based on image recognition and analysis | |
| Wang et al. | Automated defect and contaminant inspection of HVAC duct | |
| Zhang et al. | A novel image detection method for internal cracks in corn seeds in an industrial inspection line | |
| CN108073873A (en) | Human face detection and tracing system based on high-definition intelligent video camera | |
| Samsami et al. | Classification of the air quality level based on analysis of the sky images | |
| Xia et al. | Abnormal event detection method in surveillance video based on temporal CNN and sparse optical flow | |
| CN111127355A (en) | Method for finely complementing defective light flow graph and application thereof | |
| CN107992863B (en) | Multi-resolution grain insect variety visual identification method | |
| Roy et al. | AWDMC-Net: classification of adversarial weather degraded multiclass scenes using a convolution neural network | |
| Li et al. | Image quality classification algorithm based on InceptionV3 and SVM | |
| Zhou et al. | An optional threshold with SVM cloud detection algorithm and DSP implementation | |
| CN114359578A (en) | Application method and system of pest and disease damage identification intelligent terminal |
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 |