CN109444845B - Device and method for identifying coal-rock interface based on solid-state lidar imaging - Google Patents

Abstract

The invention relates to a device and a method for identifying a coal-rock interface based on solid-state laser radar, belonging to the technical field of coal-rock identification. The device comprises a plurality of laser radar modules, a signal transmission module, a data storage module, a radar imaging module, an image fusion module and an image recognition module; the plurality of laser radar modules are used for transmitting radar signals to the same area of the coal rock mine to obtain a plurality of groups of coal rock mine data information of the same area of the coal rock mine; the signal transmission module is used for transmitting the data information of the plurality of groups of coal rock ores to the data storage module; the data storage module is used for storing a plurality of groups of coal rock ore data information; the radar imaging module is used for respectively imaging each group of coal rock mine data information to obtain a plurality of coal rock texture images; the image fusion module is used for fusing the plurality of coal rock texture images to obtain fused coal rock texture images; the image recognition module is used for carrying out normalization processing on the fused coal rock texture images and recognizing the coal rock texture images to obtain a coal rock interface recognition result.

Description

Device and method for identifying coal-rock interface based on solid-state lidar imaging

technical field

The invention relates to the technical field of coal-rock identification, in particular to a device and method for identifying coal-rock interfaces based on solid-state laser radar imaging.

Background technique

Coal rock identification is to identify whether the coal rock object is a coal mine or a rock. In the process of coal production, coal rock identification technology can be widely used in the production links such as drum coal mining, tunneling, top coal caving mining, raw coal dressing and stone grinding, etc. It is of great significance to realize safe and efficient production of coal mines and comprehensive mechanized coal mining.

At this stage, there are many coal rock identification methods, such as natural ray detection method, stress pick method, infrared detection method, active power monitoring method, vibration detection method, sound detection method, dust detection method, etc. However, due to the complex and changeable coal seam geological conditions, the above methods do not have universal applicability. At the same time, due to the harsh environment of the working face and the real-time identification, these methods are not widely used in coal rock identification.

Contents of the invention

In order to solve the problems of poor applicability and poor real-time recognition of existing coal-rock identification methods, on the one hand, the present invention provides a device for identifying coal-rock interfaces based on solid-state laser radar imaging, which includes : Multiple laser radar modules, signal transmission modules, data storage modules, radar imaging modules, image fusion modules and image recognition modules;

A plurality of laser radar modules are used to transmit radar signals to the same area of the coal mine to obtain multiple sets of coal mine data information in the same area of the coal mine, wherein the laser radar module can transmit radar signals to the coal mine , and obtain the data information of the coal mine according to the reflection signal reflected by the coal mine;

The signal transmission module is used to transmit the multiple sets of coal and rock mine data information to the data storage module;

The data storage module is used to store multiple sets of coal, rock and mine data information transmitted by the signal transmission module;

The radar imaging module is used to retrieve multiple sets of coal and rock mine data information stored in the data storage module, and image each set of coal and rock mine data information to obtain the coal rock texture image corresponding to each set of coal and rock mine data information, That is, multiple coal texture images in the same area of the coal mine;

The image fusion module is used to fuse the multiple coal and rock texture images to obtain a fused coal and rock texture image; the image recognition module is used to normalize the fused coal and rock texture images , and identify the normalized image to obtain the coal-rock interface identification result.

Each of the lidar modules includes a radar signal transmitting unit, a radar reflection signal receiving unit, and a radar signal A/D conversion unit;

a radar signal transmitting unit for transmitting radar signals to the coal mine;

The radar reflection signal receiving unit is used to receive the reflection signal reflected by the coal rock mine;

The radar signal A/D conversion unit is used to perform data conversion on the reflected signal to obtain coal, rock and mine data information.

The lidar module is a solid-state lidar.

Normalize the existing coal and rock texture images, build a fully convolutional neural network model, use the normalized existing coal and rock texture images to train and test the fully convolutional neural network model, and get the training completed The full convolutional neural network model; the fully trained full convolutional neural network model is loaded into the image recognition module;

The image recognition module performs normalization processing on the fused coal and rock texture image, and inputs the normalized image into the trained fully convolutional neural network model, and the trained fully convolutional neural network model The network model outputs the coal-rock interface recognition results.

The depth of the fully convolutional neural network model after the training is five layers, which are respectively the first layer, the second layer, the third layer, the fourth layer and the fifth layer;

The first layer consists of convolutional layer C1, convolutional layer C2, and pooling layer P1. Both convolutional layer C1 and convolutional layer C2 include 64 convolution kernels of size 3*3 and a ReLU activation function; convolution Layer C1 is used to input the normalized image, the pixel size of the normalized image is 320*320*1, and the normalized image passes through all volumes of the convolutional layer C1 After the product kernel and ReLU activation function are processed, the output feature map A1, the pixel size of the feature map A1 is 318*318*64; the convolution layer C2 is used to input the feature map A1, and the feature map A1 passes through all the convolutions of the convolution layer C2 After the kernel and ReLU activation function are processed, the output feature map A2, the pixel size of the feature map A2 is 316*316*64; the pooling layer P1 is used to input the feature map A2, and divides multiple 2*2 on the feature map A2 block, and after taking the maximum value in each block, output feature map A3, the pixel size of feature map A3 is 158*158*64;

The second layer consists of convolutional layer C3, convolutional layer C4, and pooling layer P2. Both convolutional layer C3 and convolutional layer C4 include 128 convolution kernels with a size of 2*2 and a ReLU activation function; convolution Layer C3 is used to input the feature map A3. After the feature map A3 is processed by all the convolution kernels and ReLU activation functions of the convolution layer C3, the output feature map A4, the pixel size of the feature map A4 is 156*156*128; the convolution layer C4 is used to input the feature map A4. After the feature map A4 is processed by all the convolution kernels and ReLU activation functions of the convolutional layer C4, the output feature map A5, the pixel size of the feature map A5 is 154*154*128; the pooling layer P2 It is used to input the feature map A5, and divide multiple 2*2 blocks on the feature map A5, and after taking the maximum value of each block, output the feature map A6, and the pixel size of the feature map A6 is 77* 77*128;

The third layer consists of convolutional layer C5 and convolutional layer C6. Both convolutional layer C5 and convolutional layer C6 include 256 convolution kernels of size 3*3 and a ReLU activation function; convolutional layer C5 is used for input Feature map A6, after the feature map A6 is processed by all the convolution kernels and ReLU activation functions of the convolutional layer C5, the output feature map A7, the pixel size of the feature map A7 is 75*75*256; the convolutional layer C6 is used for input features Figure A7, after the feature map A7 is processed by all convolution kernels and ReLU activation functions of the convolution layer C6, the output feature map A8, the pixel size of the feature map A8 is 73*73*256;

The fourth layer consists of upsampling layer U1, convolutional layer C7 and convolutional layer C8. Upsampling layer U1 includes 256 convolution kernels with a size of 2*2. Convolutional layer C7 and convolutional layer C8 each include 128 A convolution kernel with a size of 3*3 and a ReLU activation function; the upsampling layer U1 is used to input the feature map A8, and after the feature map A8 is deconvoluted by all the convolution kernels of the upsampling layer U1, the output feature map A9 , the pixel size of the feature map A9 is 146*146*256; the convolutional layer C7 is used to input the feature map A9, and the feature map A9 is processed by all the convolution kernels and the ReLU activation function of the convolutional layer C7 to output the feature map A10, The pixel size of the feature map A10 is 144*144*128; the convolutional layer C8 is used to input the feature map A10, and the feature map A10 is processed by all the convolution kernels and the ReLU activation function of the convolutional layer C8 to output the feature map A11. The pixel size of Figure A11 is 142*142*128;

The fifth layer consists of upsampling layer U2, convolutional layer C9, convolutional layer C10 and convolutional layer C11. Upsampling layer U2 includes 128 convolution kernels with a size of 2*2, convolutional layer C9 and convolutional layer C10 includes 64 convolution kernels of size 3*3 and a ReLU activation function, convolution layer C11 includes 2 convolution kernels of size 1*1 and a ReLU activation function; upsampling layer U2 is used for input features Figure A11, after the feature map A11 is deconvoluted by all the convolution kernels of the upsampling layer U2, the output feature map A12, the pixel size of the feature map A12 is 284*284*128; the convolution layer C9 is used to input the feature map A12, after the feature map A12 is processed by all the convolution kernels and ReLU activation functions of the convolutional layer C9, the output feature map A13, the pixel size of the feature map A13 is 282*282*64; the convolutional layer C10 is used to input the feature map A13 , after the feature map A13 is processed by all the convolution kernels and ReLU activation functions of the convolutional layer C10, the output feature map A14, the pixel size of the feature map A14 is 280*280*64; the convolutional layer C11 is used to input the feature map A14, After the feature map A14 is processed by all the convolution kernels of the convolution layer C11 and the ReLU activation function, the output feature map A15, the pixel size of the feature map A15 is 280*280*2, and the output features of the feature map A15 include coal mines of coal mines interface and rock interface.

The device also includes a power supply module for supplying power to the multiple laser radar modules, the signal transmission module, the data storage module, the radar imaging module, the image fusion module and the image recognition module.

In another aspect, the present invention provides a method for identifying a coal-rock interface based on solid-state lidar imaging, the method comprising:

Multiple laser radar modules are used to transmit radar signals to the same area of the coal mine to obtain multiple sets of coal mine data information, wherein each of the laser radar modules can transmit radar signals to the coal mine, and according to the The reflection signal reflected by the mine can obtain the data information of the coal mine;

storing the multiple sets of coal and rock mine data information;

Retrieve multiple sets of coal and rock mine data information stored, and image each set of coal and rock mine data information to obtain the coal rock texture image corresponding to each set of coal and rock mine data information, that is, multiple coal and rock mine data in the same area coal rock texture image;

Fusing the plurality of coal and rock texture images to obtain a fused coal and rock texture image;

A normalization process is performed on the fused coal-rock texture image, and the normalized image is recognized to obtain a coal-rock interface recognition result.

The laser radar module can all transmit radar signals to the coal mine, and obtain coal mine data information according to the reflected signal reflected by the coal mine, including:

transmitting radar signals to the coal mine through the radar signal transmitting unit of the laser radar module;

Through the radar reflection signal receiving unit of the lidar module, the reflection signal reflected by the coal mine is received; through the radar signal A/D conversion unit of the lidar module, data conversion is performed on the reflection signal, Get coal mine data information.

The said fused coal rock texture image is normalized, and the normalized image is identified, including:

Normalize the existing coal and rock texture images, build a fully convolutional neural network model, use the normalized existing coal and rock texture images to train and test the fully convolutional neural network model, and get the training completed The fully convolutional neural network model of ;

Perform normalization processing on the fused coal texture image, and input the normalized image into the trained full convolutional neural network model, and the trained full convolutional neural network model outputs the coal rock texture image The interface recognition result.

The depth of the fully convolutional neural network model after the training is five layers, which are respectively the first layer, the second layer, the third layer, the fourth layer and the fifth layer;

The first layer consists of convolutional layer C1, convolutional layer C2, and pooling layer P1. Both convolutional layer C1 and convolutional layer C2 include 64 convolution kernels of size 3*3 and a ReLU activation function; convolution Layer C1 is used to input the normalized image, the pixel size of the normalized image is 320*320*1, and the normalized image passes through all volumes of the convolutional layer C1 After the product kernel and ReLU activation function are processed, the output feature map A1, the pixel size of the feature map A1 is 318*318*64; the convolution layer C2 is used to input the feature map A1, and the feature map A1 passes through all the convolutions of the convolution layer C2 After the kernel and ReLU activation function are processed, the output feature map A2, the pixel size of the feature map A2 is 316*316*64; the pooling layer P1 is used to input the feature map A2, and divides multiple 2*2 on the feature map A2 block, and after taking the maximum value in each block, output feature map A3, the pixel size of feature map A3 is 158*158*64;

The second layer consists of convolutional layer C3, convolutional layer C4, and pooling layer P2. Both convolutional layer C3 and convolutional layer C4 include 128 convolution kernels with a size of 2*2 and a ReLU activation function; convolution Layer C3 is used to input the feature map A3. After the feature map A3 is processed by all the convolution kernels and ReLU activation functions of the convolution layer C3, the output feature map A4, the pixel size of the feature map A4 is 156*156*128; the convolution layer C4 is used to input the feature map A4. After the feature map A4 is processed by all the convolution kernels and ReLU activation functions of the convolutional layer C4, the output feature map A5, the pixel size of the feature map A5 is 154*154*128; the pooling layer P2 It is used to input the feature map A5, and divide multiple 2*2 blocks on the feature map A5, and after taking the maximum value of each block, output the feature map A6, and the pixel size of the feature map A6 is 77* 77*128;

The third layer consists of convolutional layer C5 and convolutional layer C6. Both convolutional layer C5 and convolutional layer C6 include 256 convolution kernels of size 3*3 and a ReLU activation function; convolutional layer C5 is used for input Feature map A6, after the feature map A6 is processed by all the convolution kernels and ReLU activation functions of the convolutional layer C5, the output feature map A7, the pixel size of the feature map A7 is 75*75*256; the convolutional layer C6 is used for input features Figure A7, after the feature map A7 is processed by all convolution kernels and ReLU activation functions of the convolution layer C6, the output feature map A8, the pixel size of the feature map A8 is 73*73*256;

The fourth layer consists of upsampling layer U1, convolutional layer C7 and convolutional layer C8. Upsampling layer U1 includes 256 convolution kernels with a size of 2*2. Convolutional layer C7 and convolutional layer C8 each include 128 A convolution kernel with a size of 3*3 and a ReLU activation function; the upsampling layer U1 is used to input the feature map A8, and after the feature map A8 is deconvoluted by all the convolution kernels of the upsampling layer U1, the output feature map A9 , the pixel size of the feature map A9 is 146*146*256; the convolutional layer C7 is used to input the feature map A9, and the feature map A9 is processed by all the convolution kernels and the ReLU activation function of the convolutional layer C7 to output the feature map A10, The pixel size of the feature map A10 is 144*144*128; the convolutional layer C8 is used to input the feature map A10, and the feature map A10 is processed by all the convolution kernels and the ReLU activation function of the convolutional layer C8 to output the feature map A11. The pixel size of Figure A11 is 142*142*128;

The fifth layer consists of upsampling layer U2, convolutional layer C9, convolutional layer C10 and convolutional layer C11. Upsampling layer U2 includes 128 convolution kernels with a size of 2*2, convolutional layer C9 and convolutional layer C10 includes 64 convolution kernels of size 3*3 and a ReLU activation function, convolution layer C11 includes 2 convolution kernels of size 1*1 and a ReLU activation function; upsampling layer U2 is used for input features Figure A11, after the feature map A11 is deconvoluted by all the convolution kernels of the upsampling layer U2, the output feature map A12, the pixel size of the feature map A12 is 284*284*128; the convolution layer C9 is used to input the feature map A12, after the feature map A12 is processed by all the convolution kernels and ReLU activation functions of the convolutional layer C9, the output feature map A13, the pixel size of the feature map A13 is 282*282*64; the convolutional layer C10 is used to input the feature map A13 , after the feature map A13 is processed by all the convolution kernels and ReLU activation functions of the convolutional layer C10, the output feature map A14, the pixel size of the feature map A14 is 280*280*64; the convolutional layer C11 is used to input the feature map A14, After the feature map A14 is processed by all the convolution kernels of the convolution layer C11 and the ReLU activation function, the output feature map A15, the pixel size of the feature map A15 is 280*280*2, and the output features of the feature map A15 include coal mines of coal mines interface and rock interface.

Through the above technical solutions, compared with the prior art, the present invention has the following beneficial effects:

In the present invention, radar signals are used to detect the coal-rock interface, and the detection accuracy can reach the millimeter level, and the relative depth of the uneven surface of the coal rock can be detected. The detection process does not depend on environmental radiation, and the anti-interference ability is strong. Multiple laser radar modules are used to detect Radar signals are emitted in the same area of the coal mine to form multiple coal texture images in the same area. By fusing multiple coal texture images in the same area, the accuracy of coal mine image imaging is improved, and the full convolutional neural network is used to The network model recognizes the coal mine interface and rock interface on the fused coal and rock texture image, making the recognition result more accurate; the present invention has strong anti-interference ability in mines in complex environments, and can accurately identify coal and rock, and The operation process is simple, the applicability is good, and the distribution of coal mines and rocks can be identified in real time.

Description of drawings

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

Fig. 1 is the structural representation of the device for identifying the coal-rock interface based on solid-state lidar imaging of the present invention;

Fig. 2 is a schematic layout diagram of the lidar module of the present invention;

Fig. 3 is the full convolutional neural network structural diagram of the present invention;

Fig. 4 is a flow chart of the method for identifying coal-rock interfaces based on solid-state lidar imaging according to the present invention.

Detailed ways

The present invention is described in further detail now in conjunction with accompanying drawing. These drawings are all simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, so they only show the configurations related to the present invention.

Example 1

In order to solve the problems of poor applicability and poor real-time recognition of existing coal-rock identification methods, as shown in Figure 1, the present invention provides a device for identifying coal-rock interfaces based on solid-state laser radar imaging. The device includes: multiple laser radar modules, signal transmission modules, data storage modules, radar imaging modules, image fusion modules and image recognition modules; multiple laser radar modules are used to transmit radar signals to the same area of coal mines to obtain coal Multiple sets of coal and rock mine data information in the same area of the rock mine, among which, the laser radar module can send radar signals to the coal and rock mine, and obtain the coal and rock mine data information according to the reflected signal reflected by the coal and rock mine;

In the present invention, each laser radar module includes a radar signal transmitting unit, a radar reflection signal receiving unit, and a radar signal A/D conversion unit;

The radar signal transmitting unit is used for transmitting radar signals to coal mines;

The radar reflection signal receiving unit is used to receive the reflection signal reflected by the coal mine;

The radar signal A/D conversion unit is used for data conversion of the reflected signal to obtain coal, rock and mine data information.

Among them, the radar signal transmitting unit of each laser radar module transmits radar signals to the same area on the surface of the coal mine, and the radar signal emitted by each laser radar module will be reflected on the surface of the coal mine, and will also penetrate the coal mine and For reflection, the radar reflection signal receiving unit of each lidar module receives the reflection signal of the radar signal emitted by its own radar signal transmitting unit, and performs data conversion to obtain a set of coal, rock and mine data information. Therefore, multiple laser radar The module will get multiple sets of coal mine data information in the same area of the coal mine.

Wherein, as shown in Figure 2, the laser radar module among the present invention can be solid-state laser radar, can adopt the solid-state laser radar that model is CE-30, and a plurality of solid-state laser radars are arranged in a row and are placed in front of the coal rock mine surface , the solid-state laser radar moves linearly to generate radar signals on the coal mine surface. For example, as shown in Figure 2, a plurality of connected areas can be divided on the coal mine surface, which are respectively area 1, area 2, area 3 to area N, each laser radar module transmits radar signals to the surface of the coal mine in the order of area 1, area 2, area 3 to area N, so for the same area, multiple sets of coal mine data information can be obtained, as shown in Figure 2 Only three regions on the surface of the coal mine are shown, and the situation shown in Fig. 2 is that three laser radar modules are used to detect the surface of the coal mine in region 2 by laser.

The signal transmission module is used to transmit multiple sets of coal, rock and mine data information to the data storage module;

Wherein, the signal transmission module can perform data transmission through Ethernet.

The data storage module is used to store multiple sets of coal, rock and mine data information transmitted by the signal transmission module;

The radar imaging module is used to retrieve multiple sets of coal and rock mine data information stored in the data storage module, and image each set of coal and rock mine data information to obtain the coal rock texture image corresponding to each set of coal and rock mine data information, That is, multiple coal texture images in the same area of the coal mine;

The image fusion module is used to fuse multiple coal and rock texture images to obtain a coal and rock texture image after fusion;

The image recognition module is used to normalize the fused coal-rock texture image, and identify the normalized image to obtain the coal-rock interface recognition result, that is, to identify the coal mine interface and the rock interface, so The distribution of coal mines and rocks in the coal mine is obtained, wherein multiple groups of coal texture images in each area of the coal mine are imaged, fused and identified, and the coal rock identification results of each area of the coal mine are obtained.

In the present invention, the data storage module, the radar imaging module, the image fusion module and the image recognition module can be integrated in the upper computer, the upper computer can be a PC computer, and the solid-state laser radar can be used as the lower computer.

In the present invention, the fully convolutional neural network model can be used to identify coal and rock texture images, specifically:

Normalize the existing coal and rock texture images, construct a fully convolutional neural network model, use the normalized existing coal and rock texture images to train and test the fully convolutional neural network model, and obtain training Completed fully convolutional neural network model;

Among them, the existing coal and rock texture images are pre-collected coal and rock texture images, the coal mine interface and rock interface corresponding to the existing coal and rock texture images are also known, and the existing coal and rock texture images are normalized After processing, the existing coal and rock texture image after normalization processing can be divided into two parts, one part is used as training data, and the other part is used as test data. The training data is used to train the fully convolutional neural network model, and then the test data is used to train the full convolutional neural network model. The data is tested on the fully convolutional neural network model. If the error between the test result and the known coal mine interface and rock interface is small, the test result meets the requirements. If the test result does not meet the requirements, the training data is added to the full convolutional neural network. The model is trained until the test results meet the requirements, and the trained fully convolutional neural network model is obtained.

Load the trained fully convolutional neural network model into the image recognition module, use the image recognition module to normalize the coal and rock texture image fused by the image fusion module, and input the normalized image to the trained In the fully convolutional neural network model, the trained fully convolutional neural network model outputs the coal-rock interface recognition results, and the coal-rock recognition results include coal mine interfaces and rock interfaces.

Specifically, in the present invention, the depth of the trained fully convolutional neural network model is five layers, which are respectively the first layer, the second layer, the third layer, the fourth layer and the fifth layer, as shown in Figure 3 The structure diagram of the full convolutional neural network in the present invention; the first layer is made up of convolutional layer C1, convolutional layer C2 and pooling layer P1, and convolutional layer C1 and convolutional layer C2 all include 64 with a size of 3 *3 convolution kernel and a ReLU activation function; convolution layer C1 is used to input the normalized image, the normalized image pixel size is 320*320*1, after normalization After the image of the convolution layer C1 is convolved with all the convolution kernels and the ReLU activation function, the output feature map A1, the pixel size of the feature map A1 is 318*318*64; the convolution layer C2 is used to input the feature map A1 , after the feature map A1 is convoluted by all convolution kernels and ReLU activation functions of the convolution layer C2, the output feature map A2, the pixel size of the feature map A2 is 316*316*64; the pooling layer P1 is used for input features Figure A2, and divide multiple 2*2 blocks on the feature map A2, and after taking the maximum value in each block, output the feature map A3, the pixel size of the feature map A3 is 158*158*64;

The second layer consists of convolutional layer C3, convolutional layer C4, and pooling layer P2. Both convolutional layer C3 and convolutional layer C4 include 128 convolution kernels with a size of 2*2 and a ReLU activation function; convolution Layer C3 is used to input the feature map A3. After the feature map A3 is convoluted by all the convolution kernels and ReLU activation functions of the convolution layer C3, the output feature map A4, the pixel size of the feature map A4 is 156*156*128; The convolutional layer C4 is used to input the feature map A4. After the feature map A4 is convoluted by all the convolution kernels and the ReLU activation function of the convolutional layer C4, the output feature map A5, the pixel size of the feature map A5 is 154*154* 128; the pooling layer P2 is used to input the feature map A5, and divide a plurality of 2*2 blocks on the feature map A5, and after taking the maximum value in each block, output the feature map A6, the feature map A6 The pixel size is 77*77*128;

The third layer consists of convolutional layer C5 and convolutional layer C6. Both convolutional layer C5 and convolutional layer C6 include 256 convolution kernels of size 3*3 and a ReLU activation function; convolutional layer C5 is used for input The feature map A6, after the feature map A6 is convoluted by all the convolution kernels and the ReLU activation function of the convolution layer C5, the output feature map A7, the pixel size of the feature map A7 is 75*75*256; the convolution layer C6 uses In the input feature map A7, after the feature map A7 is convoluted by all convolution kernels and ReLU activation functions of the convolution layer C6, the output feature map A8, the pixel size of the feature map A8 is 73*73*256;

The fourth layer consists of upsampling layer U1, convolutional layer C7 and convolutional layer C8. Upsampling layer U1 includes 256 convolution kernels with a size of 2*2. Convolutional layer C7 and convolutional layer C8 each include 128 A convolution kernel with a size of 3*3 and a ReLU activation function; the upsampling layer U1 is used to input the feature map A8, and after the feature map A8 is deconvoluted by all the convolution kernels of the upsampling layer U1, the output feature map A9 , the pixel size of the feature map A9 is 146*146*256; the convolutional layer C7 is used to input the feature map A9, and the feature map A9 is convoluted by all the convolution kernels and the ReLU activation function of the convolutional layer C7, and the output feature In Figure A10, the pixel size of the feature map A10 is 144*144*128; the convolutional layer C8 is used to input the feature map A10, and the feature map A10 is convoluted by all the convolution kernels and the ReLU activation function of the convolutional layer C8. Output feature map A11, the pixel size of feature map A11 is 142*142*128;

The fifth layer consists of upsampling layer U2, convolutional layer C9, convolutional layer C10 and convolutional layer C11. Upsampling layer U2 includes 128 convolution kernels with a size of 2*2, convolutional layer C9 and convolutional layer C10 includes 64 convolution kernels of size 3*3 and a ReLU activation function, convolution layer C11 includes 2 convolution kernels of size 1*1 and a ReLU activation function; upsampling layer U2 is used for input features Figure A11, after the feature map A11 is deconvoluted by all the convolution kernels of the upsampling layer U2, the output feature map A12, the pixel size of the feature map A12 is 284*284*128; the convolution layer C9 is used to input the feature map A12, after the feature map A12 is convoluted by all convolution kernels and ReLU activation functions of the convolution layer C9, the output feature map A13, the pixel size of the feature map A13 is 282*282*64; the convolution layer C10 is used for input The feature map A13, after the feature map A13 is convoluted by all the convolution kernels and the ReLU activation function of the convolution layer C10, the output feature map A14, the pixel size of the feature map A14 is 280*280*64; the convolution layer C11 uses In the input feature map A14, after the feature map A14 is convoluted by all convolution kernels and ReLU activation functions of the convolution layer C11, the output feature map A15, the pixel size of the feature map A15 is 280*280*2, and the feature map A15 The output features for the coal-rock mine include the coal interface and the rock interface.

The device of the present invention may also include a power supply module for supplying power to multiple laser radar modules, signal transmission modules, data storage modules, radar imaging modules, image fusion modules and image recognition modules. In the present invention, the data storage module, The radar imaging module, image fusion module and image recognition module can all be integrated in the host computer.

The device in the present invention uses radar signals to detect the coal-rock interface, the detection accuracy can reach the millimeter level, and the relative depth of the uneven surface of the coal rock can be detected. The detection process does not depend on environmental radiation, and the anti-interference ability is strong. The radar module transmits radar signals to the same area of coal and rock mines to form multiple coal and rock texture images in the same area. By fusing multiple coal and rock texture images in the same area, the accuracy of coal and rock mine image imaging is improved. The convolutional neural network model recognizes the coal mine interface and rock interface on the fused coal and rock texture image, making the recognition result more accurate; the device in the present invention has strong anti-interference ability in mines in complex environments, and can accurately Coal and rock identification, and the operation process is simple, the applicability is good, and it can identify the distribution of coal mines and rocks in real time.

Example 2

The present invention provides a method for identifying the coal-rock interface based on solid-state laser radar imaging, as shown in Figure 4, the method comprising:

101. Multiple laser radar modules are used to transmit radar signals to the same area of the coal mine, and multiple sets of coal mine data information are obtained. Each laser radar module can transmit radar signals to the coal mine, and according to the coal mine The reflection signal reflected by the mine can obtain the data information of the coal mine;

Wherein, the laser radar module includes a radar signal transmitting unit, a radar reflection signal receiving unit, and a radar signal A/D conversion unit. The signal obtains coal mine data information, including:

Send radar signals to coal mines through the radar signal transmitting unit of the laser radar module;

Through the radar reflection signal receiving unit of the laser radar module, the reflection signal reflected by the coal mine is received;

Through the radar signal A/D conversion unit of the laser radar module, data conversion is performed on the reflected signal to obtain coal, rock and mine data information.

The radar signal transmitting unit of each laser radar module transmits radar signals to the same area on the surface of the coal mine, and the radar signal emitted by each laser radar module will be reflected on the surface of the coal mine, and will also penetrate the coal mine and be reflected , the radar reflection signal receiving unit of each laser radar module receives the reflection signal of the radar signal emitted by its own radar signal transmitting unit, and performs data conversion to obtain a set of coal, rock and mine data information. Therefore, multiple laser radar modules will Get multiple sets of coal and rock mine data information in the same area of the coal and rock mine.

Wherein, as shown in Figure 2, the lidar module among the present invention can be solid-state lidar, and a plurality of solid-state lidars are arranged in a row and are placed in the front of coal rock mine surface, and the mode of solid-state lidar linear movement is opposite to coal rock mine Radar signals are generated on the surface. For example, as shown in Figure 2, a plurality of connected areas can be divided on the coal mine surface, which are respectively Area 1, Area 2, Area 3 to Area N. Area 2, area 3 to area N sequentially transmit radar signals to the surface of coal and rock mines, so for the same area, multiple sets of coal and rock mine data information can be obtained.

102. Store multiple sets of coal mine data information;

103. Retrieve multiple sets of stored coal and rock mine data information, and image each set of coal and rock mine data information to obtain the coal rock texture image corresponding to each set of coal rock mine data information, that is, the coal rock mine data information in the same area Multiple coal rock texture images;

104. Fusing multiple coal and rock texture images to obtain a fused coal and rock texture image;

105. Normalize the fused coal-rock texture image, and identify the normalized image to obtain the coal-rock interface recognition result, that is, identify the coal mine interface and rock interface, thus obtaining the coal-rock interface According to the distribution of coal and rocks in the mine, multiple groups of coal texture images in each area of the coal mine are imaged, fused and identified, and the coal rock identification results of each area of the coal mine are obtained.

Wherein, normalization processing is performed on the fused coal and rock texture image, and the image after normalization processing is recognized. In the present invention, the full convolutional neural network model can be used to analyze the The image is identified, specifically:

Normalize the existing coal and rock texture images, construct a fully convolutional neural network model, use the normalized existing coal and rock texture images to train and test the fully convolutional neural network model, and obtain training Completed fully convolutional neural network model;

Among them, the existing coal and rock texture images are pre-collected coal and rock texture images, the coal mine interface and rock interface corresponding to the existing coal and rock texture images are also known, and the existing coal and rock texture images are normalized After processing, the existing coal and rock texture image after normalization processing can be divided into two parts, one part is used as training data, and the other part is used as test data. The training data is used to train the fully convolutional neural network model, and then the test data is used to train the full convolutional neural network model. The data is used to test the fully convolutional neural network model. If the error between the test result and the known coal mine interface and rock interface is small, the test result meets the requirements. The network model is trained until the test results meet the requirements, and the trained fully convolutional neural network model is obtained.

Perform normalization processing on the fused coal texture image, and input the normalized image into the trained full convolutional neural network model, and the trained full convolutional neural network model outputs the coal rock texture image Interface identification results, coal and rock identification results include coal mine interface and rock interface.

Specifically, in the present invention, the depth of the trained fully convolutional neural network model is five layers, which are respectively the first layer, the second layer, the third layer, the fourth layer and the fifth layer, as shown in Figure 3 It is the convolutional neural network structure diagram among the present invention; The first layer is made up of convolutional layer C1, convolutional layer C2 and pooling layer P1, and convolutional layer C1 and convolutional layer C2 all include 64 size is 3* 3 convolution kernel and a ReLU activation function; the convolution layer C1 is used to input the normalized image, the normalized image pixel size is 320*320*1, and the normalized image pixel size is 320*320*1, and the normalized After the image is convoluted by all the convolution kernels and ReLU activation functions of the convolutional layer C1, the output feature map A1, the pixel size of the feature map A1 is 318*318*64; the convolutional layer C2 is used to input the feature map A1, After the feature map A1 is convoluted by all the convolution kernels and ReLU activation functions of the convolution layer C2, the output feature map A2, the pixel size of the feature map A2 is 316*316*64; the pooling layer P1 is used to input the feature map A2, and divide a plurality of 2*2 blocks on the feature map A2, and after taking the maximum value of each block, output the feature map A3, the pixel size of the feature map A3 is 158*158*64;

The second layer consists of convolutional layer C3, convolutional layer C4, and pooling layer P2. Both convolutional layer C3 and convolutional layer C4 include 128 convolution kernels with a size of 2*2 and a ReLU activation function; convolution Layer C3 is used to input the feature map A3. After the feature map A3 is convoluted by all the convolution kernels and ReLU activation functions of the convolution layer C3, the output feature map A4, the pixel size of the feature map A4 is 156*156*128; The convolutional layer C4 is used to input the feature map A4. After the feature map A4 is convoluted by all the convolution kernels and the ReLU activation function of the convolutional layer C4, the output feature map A5, the pixel size of the feature map A5 is 154*154* 128; the pooling layer P2 is used to input the feature map A5, and divide a plurality of 2*2 blocks on the feature map A5, and after taking the maximum value in each block, output the feature map A6, the feature map A6 The pixel size is 77*77*128;

The third layer consists of convolutional layer C5 and convolutional layer C6. Both convolutional layer C5 and convolutional layer C6 include 256 convolution kernels of size 3*3 and a ReLU activation function; convolutional layer C5 is used for input The feature map A6, after the feature map A6 is convoluted by all the convolution kernels and the ReLU activation function of the convolution layer C5, the output feature map A7, the pixel size of the feature map A7 is 75*75*256; the convolution layer C6 uses In the input feature map A7, after the feature map A7 is convoluted by all convolution kernels and ReLU activation functions of the convolution layer C6, the output feature map A8, the pixel size of the feature map A8 is 73*73*256;

The fourth layer consists of upsampling layer U1, convolutional layer C7 and convolutional layer C8. Upsampling layer U1 includes 256 convolution kernels with a size of 2*2. Convolutional layer C7 and convolutional layer C8 each include 128 A convolution kernel with a size of 3*3 and a ReLU activation function; the upsampling layer U1 is used to input the feature map A8, and after the feature map A8 is deconvoluted by all the convolution kernels of the upsampling layer U1, the output feature map A9 , the pixel size of the feature map A9 is 146*146*256; the convolutional layer C7 is used to input the feature map A9, and the feature map A9 is convoluted by all the convolution kernels and the ReLU activation function of the convolutional layer C7, and the output feature In Figure A10, the pixel size of the feature map A10 is 144*144*128; the convolutional layer C8 is used to input the feature map A10, and the feature map A10 is convoluted by all the convolution kernels and the ReLU activation function of the convolutional layer C8. Output feature map A11, the pixel size of feature map A11 is 142*142*128;

The fifth layer consists of upsampling layer U2, convolutional layer C9, convolutional layer C10 and convolutional layer C11. Upsampling layer U2 includes 128 convolution kernels with a size of 2*2, convolutional layer C9 and convolutional layer C10 includes 64 convolution kernels of size 3*3 and a ReLU activation function, convolution layer C11 includes 2 convolution kernels of size 1*1 and a ReLU activation function; upsampling layer U2 is used for input features Figure A11, after the feature map A11 is deconvoluted by all the convolution kernels of the upsampling layer U2, the output feature map A12, the pixel size of the feature map A12 is 284*284*128; the convolution layer C9 is used to input the feature map A12, after the feature map A12 is convoluted by all convolution kernels and ReLU activation functions of the convolution layer C9, the output feature map A13, the pixel size of the feature map A13 is 282*282*64; the convolution layer C10 is used for input The feature map A13, after the feature map A13 is convoluted by all the convolution kernels and the ReLU activation function of the convolution layer C10, the output feature map A14, the pixel size of the feature map A14 is 280*280*64; the convolution layer C11 uses In the input feature map A14, after the feature map A14 is convoluted by all convolution kernels and ReLU activation functions of the convolution layer C11, the output feature map A15, the pixel size of the feature map A15 is 280*280*2, and the feature map A15 The output features for the coal mine include the coal interface and the rock interface.

The method in the present invention uses radar signals to detect the coal-rock interface, the detection accuracy can reach the millimeter level, and the relative depth of the uneven surface of the coal rock can be detected. The detection process does not depend on environmental radiation, and the anti-interference ability is strong. Multiple lasers are used The radar module transmits radar signals to the same area of coal and rock mines to form multiple coal and rock texture images in the same area. By fusing multiple coal and rock texture images in the same area, the accuracy of coal and rock mine image imaging is improved. The convolutional neural network model recognizes the coal mine interface and the rock interface on the fused coal and rock texture image, making the identification result more accurate; the method in the present invention has strong anti-interference ability in mines in complex environments, and can Coal and rock identification, and the operation process is simple, the applicability is good, and it can identify the distribution of coal mines and rocks in real time.

Inspired by the above-mentioned ideal embodiment according to the present invention, through the above-mentioned description content, relevant workers can make various changes and modifications within the scope of not departing from the technical idea of the present invention. The technical scope of the present invention is not limited to the content in the specification, but must be determined according to the scope of the claims.

Claims (6)

1. Device based on solid-state laser radar reaches carries out discernment to coal petrography interface, its characterized in that: the device comprises a plurality of laser radar modules, a signal transmission module, a data storage module, a radar imaging module, an image fusion module and an image recognition module;

the system comprises a plurality of laser radar modules, a plurality of data acquisition module and a data acquisition module, wherein the laser radar modules are used for transmitting radar signals to the same area of the coal rock and obtaining a plurality of groups of coal rock and mineral data information of the same area of the coal rock and mineral, and can transmit radar signals to the coal rock and mineral and obtain the coal rock and mineral data information according to the reflected signals reflected by the coal rock and mineral;

the signal transmission module is used for transmitting the multiple groups of coal rock mine data information to the data storage module;

the data storage module is used for storing the multiple groups of coal rock mine data information transmitted by the signal transmission module;

the radar imaging module is used for calling a plurality of groups of coal rock ore data information stored by the data storage module, and respectively imaging each group of coal rock ore data information to obtain a coal rock texture image corresponding to each group of coal rock ore data information, namely a plurality of coal rock texture images of the same area of the coal rock ore;

the image fusion module is used for fusing the plurality of coal rock texture images to obtain a fused coal rock texture image;

The image recognition module is used for carrying out normalization processing on the fused coal rock texture image, recognizing the normalized image and obtaining a coal rock interface recognition result;

carrying out normalization processing on the existing coal-rock texture image, constructing a full-convolution neural network model, and training and testing the full-convolution neural network model by utilizing the normalized existing coal-rock texture image to obtain a trained full-convolution neural network model; loading the trained full convolution neural network model into the image recognition module;

the image recognition module performs normalization processing on the fused coal-rock texture image, inputs the normalized image into a trained full-convolution neural network model, and outputs a coal-rock interface recognition result;

the depth of the trained full convolution neural network model is five layers, namely a first layer, a second layer, a third layer, a fourth layer and a fifth layer;

the first layer consists of a convolution layer C1, a convolution layer C2 and a pooling layer P1, wherein the convolution layer C1 and the convolution layer C2 comprise 64 convolution kernels with the size of 3*3 and a ReLU activation function; the convolution layer C1 is configured to input the normalized image, where the pixel size of the normalized image is 320×320×1, and after all convolution kernels and ReLU activation functions of the normalized image are processed by the convolution layer C1, a feature map A1 is output, and the pixel size of the feature map A1 is 318×318×64; the convolution layer C2 is used for inputting the feature map A1, and after the feature map A1 is processed by all convolution kernels and ReLU activation functions of the convolution layer C2, the feature map A2 is output, and the pixel size of the feature map A2 is 316×316×64; the pooling layer P1 is configured to input a feature map A2, divide a plurality of blocks of 2×2 on the feature map A2, and output a feature map A3 after taking a maximum value from each block, where a pixel size of the feature map A3 is 158×158×64;

The second layer consists of a convolution layer C3, a convolution layer C4 and a pooling layer P2, wherein the convolution layer C3 and the convolution layer C4 comprise 128 convolution kernels with the size of 2 x 2 and a ReLU activation function; the convolution layer C3 is used for inputting a feature map A3, and after the feature map A3 is processed by all convolution kernels and ReLU activation functions of the convolution layer C3, a feature map A4 is output, and the pixel size of the feature map A4 is 156×156×128; the convolution layer C4 is used for inputting a feature map A4, and after the feature map A4 is processed by all convolution kernels and ReLU activation functions of the convolution layer C4, a feature map A5 is output, and the pixel size of the feature map A5 is 154×128; the pooling layer P2 is configured to input a feature map A5, divide a plurality of blocks of 2×2 on the feature map A5, and output a feature map A6 after taking a maximum value from each block, where a pixel size of the feature map A6 is 77×77×128;

the third layer consists of a convolution layer C5 and a convolution layer C6, wherein the convolution layer C5 and the convolution layer C6 comprise 256 convolution kernels with the size of 3*3 and a ReLU activation function; the convolution layer C5 is used for inputting a feature map A6, and after the feature map A6 is processed by all convolution kernels and ReLU activation functions of the convolution layer C5, a feature map A7 is output, and the pixel size of the feature map A7 is 75 x 256; the convolution layer C6 is used for inputting a feature map A7, and after the feature map A7 is processed by all convolution kernels and ReLU activation functions of the convolution layer C6, a feature map A8 is output, and the pixel size of the feature map A8 is 73 x 256;

The fourth layer is composed of an up-sampling layer U1, a convolution layer C7 and a convolution layer C8, wherein the up-sampling layer U1 comprises 256 convolution kernels with the size of 2 x 2, and the convolution layer C7 and the convolution layer C8 comprise 128 convolution kernels with the size of 3*3 and a ReLU activation function; the up-sampling layer U1 is used for inputting a feature map A8, and after the feature map A8 is subjected to deconvolution processing by all convolution kernels of the up-sampling layer U1, a feature map A9 is output, wherein the pixel size of the feature map A9 is 146 x 256; the convolution layer C7 is used for inputting a feature map A9, and after the feature map A9 is processed by all convolution kernels and ReLU activation functions of the convolution layer C7, a feature map a10 is output, and the pixel size of the feature map a10 is 144×144×128; the convolution layer C8 is used for inputting a feature map a10, and after the feature map a10 is processed by all convolution kernels and ReLU activation functions of the convolution layer C8, a feature map a11 is output, and the pixel size of the feature map a11 is 142×142×128;

the fifth layer is composed of an up-sampling layer U2, a convolution layer C9, a convolution layer C10 and a convolution layer C11, wherein the up-sampling layer U2 comprises 128 convolution kernels with the size of 2 x 2, the convolution layer C9 and the convolution layer C10 comprise 64 convolution kernels with the size of 3*3 and a ReLU activation function, and the convolution layer C11 comprises 2 convolution kernels with the size of 1*1 and a ReLU activation function; the up-sampling layer U2 is configured to input a feature map a11, where the feature map a11 is deconvoluted by all convolution kernels of the up-sampling layer U2, and then output a feature map a12, where a pixel size of the feature map a12 is 284×284×128; the convolution layer C9 is used for inputting a feature map a12, and after the feature map a12 is processed by all convolution kernels and ReLU activation functions of the convolution layer C9, a feature map a13 is output, and the pixel size of the feature map a13 is 282×282×64; the convolution layer C10 is used for inputting a feature map a13, and after the feature map a13 is processed by all convolution kernels and ReLU activation functions of the convolution layer C10, outputting a feature map a14, wherein the pixel size of the feature map a14 is 280×280×64; the convolution layer C11 is used for inputting the feature map a14, after the feature map a14 is processed by all convolution kernels and ReLU activation functions of the convolution layer C11, outputting a feature map a15, wherein the pixel size of the feature map a15 is 280×280×2, and the output features of the feature map 15 comprise coal mine interfaces and rock interfaces of coal and rock ores.

2. The device for identifying a coal-rock interface based on solid state laser radar according to claim 1, wherein: each laser radar module comprises a radar signal transmitting unit, a radar reflection signal receiving unit and a radar signal A/D conversion unit;

the radar signal transmitting unit is used for transmitting radar signals to the coal rock mine;

the radar reflected signal receiving unit is used for receiving reflected signals reflected by the coal rock mine;

and the radar signal A/D conversion unit is used for carrying out data conversion on the reflected signals to obtain coal rock mine data information.

3. The device for identifying a coal-rock interface based on solid state laser radar according to claim 1, wherein: the laser radar module is a solid-state laser radar.

4. The device for identifying a coal-rock interface based on solid state laser radar according to claim 1, wherein: the device also comprises a power supply module, which is used for supplying power to the laser radar modules, the signal transmission module, the data storage module, the radar imaging module, the image fusion module and the image recognition module.

5. The method for identifying the coal-rock interface based on the solid-state laser radar is characterized by comprising the following steps of: the method comprises the following steps:

transmitting radar signals to the same area of the coal rock mine by adopting a plurality of laser radar modules to obtain a plurality of groups of coal rock mine data information, wherein each laser radar module can transmit radar signals to the coal rock mine and obtain the coal rock mine data information according to the reflected signals reflected by the coal rock mine;

storing the multiple groups of coal rock ore data information;

retrieving stored multiple groups of coal rock ore data information, and respectively imaging each group of coal rock ore data information to obtain a coal rock texture image corresponding to each group of coal rock ore data information, namely multiple coal rock texture images of the same area of the coal rock ore;

fusing the plurality of coal rock texture images to obtain a fused coal rock texture image;

normalizing the fused coal rock texture image, and identifying the normalized image to obtain a coal rock interface identification result; the normalizing processing is carried out on the fused coal rock texture image, and the normalized image is identified, and the method comprises the following steps:

carrying out normalization processing on the existing coal-rock texture image, constructing a full-convolution neural network model, and training and testing the full-convolution neural network model by utilizing the normalized existing coal-rock texture image to obtain a trained full-convolution neural network model;

Normalizing the fused coal-rock texture image, inputting the normalized image into a trained full-convolution neural network model, and outputting a coal-rock interface identification result by the trained full-convolution neural network model;

the depth of the trained full convolution neural network model is five layers, namely a first layer, a second layer, a third layer, a fourth layer and a fifth layer;

the first layer consists of a convolution layer C1, a convolution layer C2 and a pooling layer P1, wherein the convolution layer C1 and the convolution layer C2 comprise 64 convolution kernels with the size of 3*3 and a ReLU activation function; the convolution layer C1 is configured to input the normalized image, where the pixel size of the normalized image is 320×320×1, and after all convolution kernels and ReLU activation functions of the normalized image are processed by the convolution layer C1, a feature map A1 is output, and the pixel size of the feature map A1 is 318×318×64; the convolution layer C2 is used for inputting the feature map A1, and after the feature map A1 is processed by all convolution kernels and ReLU activation functions of the convolution layer C2, the feature map A2 is output, and the pixel size of the feature map A2 is 316×316×64; the pooling layer P1 is configured to input a feature map A2, divide a plurality of blocks of 2×2 on the feature map A2, and output a feature map A3 after taking a maximum value from each block, where a pixel size of the feature map A3 is 158×158×64;

The second layer consists of a convolution layer C3, a convolution layer C4 and a pooling layer P2, wherein the convolution layer C3 and the convolution layer C4 comprise 128 convolution kernels with the size of 2 x 2 and a ReLU activation function; the convolution layer C3 is used for inputting a feature map A3, and after the feature map A3 is processed by all convolution kernels and ReLU activation functions of the convolution layer C3, a feature map A4 is output, and the pixel size of the feature map A4 is 156×156×128; the convolution layer C4 is used for inputting a feature map A4, and after the feature map A4 is processed by all convolution kernels and ReLU activation functions of the convolution layer C4, a feature map A5 is output, and the pixel size of the feature map A5 is 154×128; the pooling layer P2 is configured to input a feature map A5, divide a plurality of blocks of 2×2 on the feature map A5, and output a feature map A6 after taking a maximum value from each block, where a pixel size of the feature map A6 is 77×77×128;

the third layer consists of a convolution layer C5 and a convolution layer C6, wherein the convolution layer C5 and the convolution layer C6 comprise 256 convolution kernels with the size of 3*3 and a ReLU activation function; the convolution layer C5 is used for inputting a feature map A6, and after the feature map A6 is processed by all convolution kernels and ReLU activation functions of the convolution layer C5, a feature map A7 is output, and the pixel size of the feature map A7 is 75 x 256; the convolution layer C6 is used for inputting a feature map A7, and after the feature map A7 is processed by all convolution kernels and ReLU activation functions of the convolution layer C6, a feature map A8 is output, and the pixel size of the feature map A8 is 73 x 256;

The fourth layer is composed of an up-sampling layer U1, a convolution layer C7 and a convolution layer C8, wherein the up-sampling layer U1 comprises 256 convolution kernels with the size of 2 x 2, and the convolution layer C7 and the convolution layer C8 comprise 128 convolution kernels with the size of 3*3 and a ReLU activation function; the up-sampling layer U1 is used for inputting a feature map A8, and after the feature map A8 is subjected to deconvolution processing by all convolution kernels of the up-sampling layer U1, a feature map A9 is output, wherein the pixel size of the feature map A9 is 146 x 256; the convolution layer C7 is used for inputting a feature map A9, and after the feature map A9 is processed by all convolution kernels and ReLU activation functions of the convolution layer C7, a feature map a10 is output, and the pixel size of the feature map a10 is 144×144×128; the convolution layer C8 is used for inputting a feature map a10, and after the feature map a10 is processed by all convolution kernels and ReLU activation functions of the convolution layer C8, a feature map a11 is output, and the pixel size of the feature map a11 is 142×142×128;

the fifth layer is composed of an up-sampling layer U2, a convolution layer C9, a convolution layer C10 and a convolution layer C11, wherein the up-sampling layer U2 comprises 128 convolution kernels with the size of 2 x 2, the convolution layer C9 and the convolution layer C10 comprise 64 convolution kernels with the size of 3*3 and a ReLU activation function, and the convolution layer C11 comprises 2 convolution kernels with the size of 1*1 and a ReLU activation function; the up-sampling layer U2 is configured to input a feature map a11, where the feature map a11 is deconvoluted by all convolution kernels of the up-sampling layer U2, and then output a feature map a12, where a pixel size of the feature map a12 is 284×284×128; the convolution layer C9 is used for inputting a feature map a12, and after the feature map a12 is processed by all convolution kernels and ReLU activation functions of the convolution layer C9, a feature map a13 is output, and the pixel size of the feature map a13 is 282×282×64; the convolution layer C10 is used for inputting a feature map a13, and after the feature map a13 is processed by all convolution kernels and ReLU activation functions of the convolution layer C10, outputting a feature map a14, wherein the pixel size of the feature map a14 is 280×280×64; the convolution layer C11 is used for inputting the feature map a14, after the feature map a14 is processed by all convolution kernels and ReLU activation functions of the convolution layer C11, outputting a feature map a15, wherein the pixel size of the feature map a15 is 280×280×2, and the output features of the feature map 15 comprise coal mine interfaces and rock interfaces of coal and rock ores.

6. The method for identifying a coal-rock interface based on solid state laser radar according to claim 5, wherein: the laser radar module can both transmit radar signals to coal rock ore, and obtain coal rock ore data information according to reflected signals reflected by the coal rock ore, and comprises:

transmitting radar signals to the coal rock ore through a radar signal transmitting unit of the laser radar module;

receiving a reflected signal reflected by the coal rock ore through a radar reflected signal receiving unit of the laser radar module;

and performing data conversion on the reflected signals through a radar signal A/D conversion unit of the laser radar module to obtain coal rock data information.

Images