CN113159078B

CN113159078B - Image data identification system and method based on neural network

Info

Publication number: CN113159078B
Application number: CN202110623007.9A
Authority: CN
Inventors: 涂建刚; 鞠进军; 徐成; 余晓凡; 张秀丽
Original assignee: Army Engineering University of PLA
Current assignee: Army Engineering University of PLA
Priority date: 2021-06-04
Filing date: 2021-06-04
Publication date: 2022-10-04
Anticipated expiration: 2041-06-04
Also published as: CN113159078A

Abstract

The invention discloses an image data identification system and method based on a neural network, which comprises the following steps: the method comprises the following steps: data input, namely preprocessing the data after the data input is finished, processing the data into more obvious data to form recognition or training data, and then performing normalization processing on the image; step two: designing a convolution calculation layer, dividing an original image into three channels of R, G and B, decomposing one image into output layers of 9 × 6 × 3, and creating three convolutions for the output layers to generate three convolution kernels; step three: constructing an excitation function, and step four: designing a pooling layer, and step five: and designing a full connection layer. The image data identification system and method based on the neural network can identify the action characteristics of the transport vehicle with the pontoon bridge during transportation, extract the characteristic data, compare the characteristic data with a standard action characteristic library, analyze the equipment operation adaptability and the man-machine adaptability, and exert the effective degree of the equipment in the application process.

Description

Image data identification system and method based on neural network

Technical Field

The invention relates to the field of image recognition, in particular to an image data recognition system and method based on a neural network.

Background

In the existing military training, the boat bridge equipment is needed to overcome river obstacles in the military, and when the boat bridge transport vehicle is used for transporting on the boat bridge, the following problems exist: firstly, when the pontoon bridge equipment is loaded and unloaded, the pontoon bridge equipment is generally supported by the support rods of the pontoon bridge carrier, and when the pontoon bridge carrier is parked, the support rods are greatly stressed by impact force, so that the support rods are easily damaged, and the service life of the pontoon bridge carrier is influenced; meanwhile, before the boat bridge equipment works, the parking position needs to be determined, but because a plurality of boat bridge transport vehicles are transported on the surface of the boat bridge, the surface of the boat bridge is uneven, the parking position of the transport vehicles cannot be guaranteed to be a horizontal road section, and the support is unstable due to the fact that the support rods are used for supporting independently; if a plurality of supporting structures are used, the structure cannot be kept stable on uneven ground, the occupied area is large, and the structure is inconvenient to operate and carry when an emergency task occurs; meanwhile, in the daily military training process, when the pontoon bridge is equipped for operation, the action characteristics of the equipment operation cannot be identified, and the characteristic data is extracted and compared with a standard action characteristic library.

Therefore, the image data identification system and method based on the neural network can help the boat bridge equipment to find a stable ground when the boat bridge equipment is not found, the boat bridge equipment can conveniently play a role in supporting by adjusting the length of the auxiliary rod, the boat bridge equipment is convenient to carry, meanwhile, the damage to the supporting rod of the carrier vehicle is reduced, the service life of the carrier vehicle is prolonged, the boat bridge transport vehicle has good stability, meanwhile, the motion characteristics of the transport equipment can be identified when the boat bridge transport vehicle is parked, the characteristic data can be extracted and compared with a standard motion characteristic library, the operation adaptability and the man-machine adaptability of the equipment are analyzed, and the effective degree of the equipment in the application process is exerted.

Disclosure of Invention

The invention mainly aims to provide an image data identification system and method based on a neural network, which can effectively solve the problems in the background technology.

In order to achieve the purpose, the invention adopts the technical scheme that:

an image data identification system and method based on a neural network comprises the following steps:

the method comprises the following steps: data input, namely preprocessing the data after the data input is finished, processing the data into more obvious data to form recognition or training data, and then performing normalization processing on the image;

step two: designing a convolution calculation layer, dividing an original image into three channels of R, G and B, decomposing one image into output layers of 9 × 6 × 3, and creating three convolutions for the output layers to generate three convolution kernels;

step three: constructing an excitation function, and performing nonlinear mapping by using a LeakyRelu function;

step four: designing a pooling layer, selecting 2 x 2 in the system, and adopting maximum pooling, namely taking the maximum value of 4 points;

step five: and the full connection layer is designed, and the full connection layer utilizes the high-level characteristics to divide the input images into different categories according to the high-level characteristics which are output by the convolution layer and the pooling layer and represent the input images.

Including first set of splice block, second cover splice block and connecting rod, install the connecting rod between first set of splice block and the second cover splice block, the bracing piece is all installed at the automobile body rear to second cover splice block and first set of splice block, the surface at one side bracing piece is installed to first set of splice block, the surface at the opposite side bracing piece is installed to the second cover splice block, the mount pad is still installed in the front of connecting rod, first infrared camera and the infrared camera of second are installed respectively to the top of mount pad, the bottom of mount pad is equipped with the slide rail, the sliding surface of slide rail is connected with two installation pieces, and first camera and second camera are installed respectively to the surface of two installation pieces, first camera and second camera are in the bottom of mount pad.

Furthermore, first camera and second camera mutually support and constitute binocular identification system, first camera and second camera detect the bracing piece of automobile body both sides respectively, first camera and second camera are connected with the system of making level.

Further, the surface mounting of first cover joint block has first auxiliary rod, the surface mounting of second cover joint block has the second auxiliary rod, first cover joint block and second cover joint block are inside the dwang is all installed on the surface of bracing piece, first mounting bracket is installed at one side top of dwang, dwang one side mid-mounting has second comb shape draw-in groove.

Further, first mounting bracket top is half-circular arc, the runner is installed to the both sides of first mounting bracket, half-circular arc mid-mounting has the movable rod, the surface cover of movable rod is equipped with the sliding block, the dead lever is installed to the bottom of sliding block, the connecting rod is installed to the other end of dead lever.

Further, the other end of first auxiliary rod and second auxiliary rod all is equipped with the second mounting bracket, the cooperation piece is installed to the other end of second mounting bracket, the cuboid opening is seted up at the middle part of cooperation piece, the fluting has all been seted up to the both sides of cooperation piece, two flutings have been seted up respectively at the both ends of connecting rod, the movable pulley is installed at the both ends of connecting rod, the movable pulley is in the grooved outside of cooperation piece, first comb shape draw-in groove is installed to the bottom of second mounting bracket, first comb shape notch has been seted up on the surface of first comb shape draw-in groove, second comb shape notch has been seted up on the surface of second comb shape draw-in groove, the length of first comb shape draw-in groove is greater than the length of second comb shape draw-in groove, the first comb shape notch and the second comb shape draw-in groove on first comb shape draw-in groove and its surface and the second comb shape draw-in groove and the second comb shape notch size on surface match each other.

Furthermore, the leveling system comprises a first camera, a second camera and a positioning device, the leveling system shoots pictures through the first camera and the second camera and extracts key data of the pictures, the leveling system works in the process that the first camera and the second camera record images, the ranges of the images of the first camera and the second camera are circular rings with the radius of one meter formed by taking two support rods as circle centers respectively, and the leveling system adopts an image data identification method based on a convolutional neural network.

Furthermore, the leveling system records a flat ground picture in advance, inputs the flat ground picture through an input layer, trains data, extracts key characteristic data, classifies a neural network, repeats the training data until the accuracy is qualified if the identification accuracy is not good, records and captures an image captured by taking a support rod as a circle as a radius when the leveling system is actually used, inputs the image, identifies the data after inputting the image to the input layer, transmits the data to the trained convolutional layer, transmits the data to the trained neural network, outputs an identification result, judges the flat ground picture by sound or light spots or the combination of the sound and the light spots through a positioning device.

Further, binocular identification system internally mounted has pressure deformation fitting procedure, enough detects the pressure variation of bracing piece through first camera and second camera, first infrared camera and second infrared camera can carry out the record to the operation of pontoon bridge transport vechicle goods.

The use method of the system is as follows:

step a: the first sleeve connecting block and the second sleeve connecting block are respectively sleeved on the surfaces of the two supporting rods at the rear part of the vehicle body, so that the mounting is stable, the binocular recognition system records images of the environment, the accurate position of the images is conveniently judged, the leveling system searches for flat ground, the positioning device positions a proper place, the first auxiliary rod and the second auxiliary rod are placed, and the supporting rods are reinforced;

step b: if no proper position is found, the supporting rods are supported and fixed through the first auxiliary rod and the second auxiliary rod, when the ground environment is uneven, the extending lengths of the first auxiliary rod and the second auxiliary rod are respectively adjusted to be respectively positioned on the same plane with the corresponding supporting rods, one side of the first mounting frame is rotated through the rotating wheel, after the length of the proper first auxiliary rod or the proper second auxiliary rod is determined, the connecting rod slides in the groove, so that the fixed rod slides on the surface of the movable rod, the second mounting frame is moved, and after the movement, the first comb-shaped clamping groove and the second comb-shaped clamping groove are mutually clamped to be stably connected;

step c: when the pontoon bridge equipment carries out the operation, can produce the recoil to the bracing piece, cause the damage to the bracing piece, catch the deformation of bracing piece respectively through two mesh identification systems, simultaneously, can fit the pressure transformation curve, discern the loss of bracing piece, simultaneously, first infrared camera and second infrared camera can be to the goods traffic situation record in the pontoon bridge transport vechicle.

Compared with the prior art, the invention has the following beneficial effects:

1. the original supporting rod is conveniently supported by the first auxiliary rod and the second auxiliary rod, the boat bridge carrier vehicle is stabilized when being parked, the impact force damage to the boat bridge carrier vehicle is reduced, the supporting rod is supported and fixed by the first auxiliary rod and the second auxiliary rod, when the ground environment is uneven, the extending lengths of the first auxiliary rod and the second auxiliary rod are respectively adjusted to be in a plane with the corresponding supporting rod, one side of the first mounting frame is rotated by a rotating wheel, after the length of the proper first auxiliary rod or the proper second auxiliary rod is determined, the connecting rod slides in the groove, so that the fixed rod slides on the surface of the movable rod, the second mounting frame is moved, after the first mounting frame and the second mounting frame are moved, the first comb-shaped clamping groove and the second comb-shaped clamping groove are mutually clamped, the first auxiliary rod and the second auxiliary rod are stably connected, the service life of the original supporting rod is conveniently prolonged, meanwhile, the first auxiliary rod and the second auxiliary rod can be conveniently adjusted, when the carrier vehicle is in an uneven ground section, the first auxiliary rod and the supporting rod are more stable than the supporting rod, the supporting rod is more convenient for carrying of a plurality of supporting structures, and the supporting rod, and the emergency task is more suitable for carrying;

2. the vehicle can be helped to find a relatively stable ground through the arranged binocular recognition system, good external conditions are provided for the vehicle in an effort, the first sleeve connecting block and the second sleeve connecting block are respectively sleeved on the surfaces of the two supporting rods at the rear part of the vehicle body, so that the vehicle is stably installed, the binocular recognition system records images when the vehicle runs, the accurate position of the images is judged, the leveling system finds a flat ground, the positioning device positions a proper place, the first auxiliary rod and the second auxiliary rod are placed, and the supporting rods are reinforced;

3. through the binocular identification system that is equipped with, first infrared camera and the infrared deformation of second camera can be caught the bracing piece, utilize two cameras of first camera and second camera to make the distance of judgement more accurate, when the pontoon bridge equipment loads the uninstallation, carry out the fitting of pressure curve to its pressure that receives, can detect the pressure that the bracing piece received and the loss of bracing piece in the operation many times through the contrast between the pressure, the loss of discernment bracing piece, the top at the mount pad is installed to first infrared camera and second infrared camera simultaneously, can carry out the record to the goods of pontoon bridge transport vechicle loading and unloading.

Drawings

In order to more clearly illustrate the technical solutions of the present invention, the drawings required to be used in the technical description of the present invention will be briefly introduced below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings may be obtained according to the drawings without inventive labor.

FIG. 1 is a schematic diagram of a convolutional neural network architecture of a neural network-based image data recognition system and method according to the present invention;

FIG. 2 is a schematic diagram of an overall structure of a neural network-based image data recognition system and method according to the present invention;

FIG. 3 is a schematic diagram of an internal structure of a first nesting block of the image data recognition system and method based on neural network of the present invention;

FIG. 4 is a schematic structural diagram of a first mounting frame of the image data recognition system and method based on neural network according to the present invention;

FIG. 5 is a schematic structural diagram of a second mounting frame of the image data recognition system and method based on neural network according to the present invention;

FIG. 6 is a schematic structural diagram of a mounting base of the image data recognition system and method based on neural network according to the present invention;

FIG. 7 is a schematic diagram of a convolutional neural network recognition framework for a neural network-based image data recognition system and method of the present invention;

FIG. 8 is a schematic diagram of local and global features of a vehicle for a neural network based image data recognition system and method of the present invention;

FIG. 9 is a schematic diagram of image data labeling of an image data recognition system and method based on neural networks according to the present invention;

FIG. 10 is a diagram illustrating an automatic classification and recognition process of the image data recognition system and method based on neural network according to the present invention.

In the figure: 1. a vehicle body; 2. a first nesting block; 3. a second sleeving and connecting block; 4. a connecting rod; 5. a mounting base; 6. a support bar; 7. a first auxiliary lever; 8. a second auxiliary lever; 9. rotating the rod; 10. a first mounting bracket; 11. a second mounting bracket; 12. a movable rod; 13. a slider; 14. fixing the rod; 15. a first comb-shaped slot; 16. a second comb-shaped slot; 17. a first comb-shaped notch; 18. a second comb notch; 19. a matching block; 20. grooving; 21. a sliding wheel; 22. a connecting rod; 23. a slide rail; 24. mounting a block; 25. a first camera; 26. a second camera; 27. a first infrared camera; 28. and a second infrared camera.

Detailed Description

The present invention will be further described with reference to the following detailed description, wherein the drawings are for illustrative purposes only and are not intended to be limiting of the present patent, wherein certain elements may be omitted, enlarged or reduced in size to better illustrate the detailed description, and not to represent actual dimensions, and wherein certain well-known structures and descriptions may be omitted from the drawings so that those skilled in the art can understand that based on the detailed description of the present invention, all other detailed descriptions that may be obtained by those skilled in the art without making any creative effort may be within the scope of the present invention.

Example 1

As shown in fig. 2-6, an image data recognition system and method based on a neural network, including first set of joint block (2), second set of joint block (3) and connecting rod (4), install connecting rod (4) between first set of joint block (2) and second set of joint block (3), second set of joint block (3) and first set of joint block (2) are all installed at automobile body (1) rear, bracing piece (6) are all installed to the bottom both sides of automobile body (1), the surface at one side bracing piece (6) is installed to first set of joint block (2), the surface at opposite side bracing piece (6) is installed to second set of joint block (3), mount pad (5) is still installed on the front of connecting rod (4), first infrared camera (27) and second infrared camera (28) are installed respectively to the top of mount pad (5), the bottom of mount pad (5) is equipped with slide rail (23), the surface sliding connection of slide rail (23) has two mount blocks (24), first camera (25) and second camera (26) are installed respectively to the surface of two mount pads (24), first camera (25) and second camera (26) are in the mount pad (5) bottom.

First camera (25) and second camera (26) mutually support and constitute binocular identification system, and first camera (25) and second camera (26) detect bracing piece (6) of automobile body (1) both sides respectively, and first camera (25) and second camera (26) are connected with the system of making level.

The surface mounting of first suit piece (2) has first auxiliary rod (7), and the surface mounting of second suit piece has second auxiliary rod (8), and dwang (9) are all installed on the surface of the inside bracing piece (6) of first suit piece (2) and second suit piece (3), and first mounting bracket (10) are installed at one side top of dwang (9), and dwang (9) one side mid-mounting has second comb shape draw-in groove (16).

First mounting bracket (10) top is half-circular arc, and the runner is installed to the both sides of first mounting bracket (10), and half-circular arc mid-mounting has movable rod (12), and the surface cover of movable rod (12) is equipped with sliding block (13), and dead lever (14) are installed to the bottom of sliding block (13), and connecting rod (22) are installed to the other end of dead lever (14).

The other end of first auxiliary rod (7) and second auxiliary rod (8) all is equipped with second mounting bracket (11), cooperation piece (19) are installed to the other end of second mounting bracket (11), the cuboid opening is seted up at the middle part of cooperation piece (19), fluting (20) have all been seted up to the both sides of cooperation piece (19), two fluting (20) have been seted up respectively at the both ends of connecting rod (22), movable pulley (21) are installed at the both ends of connecting rod (22), movable pulley (21) are in the outside of fluting (20) of cooperation piece (19), first comb shape draw-in groove (15) are installed to the bottom of second mounting bracket (11), first comb shape notch (17) has been seted up on the surface of first comb shape draw-in groove (15), second comb shape notch (18) has been seted up on the surface of second comb shape draw-in groove (16), the length of first comb shape draw-in groove (15) is greater than the length of second comb shape draw-in groove (16), first comb shape notch (17) and the second comb shape draw-in groove (18) size of first comb shape draw-in groove (15) and its surface match each other.

By adopting the technical scheme: the original supporting rod (6) is conveniently supported by the first auxiliary rod (7) and the second auxiliary rod (8), when the boat bridge transport vehicle is parked, the boat bridge transport vehicle is stabilized, the impact force damage to the boat bridge transport vehicle is reduced, the support rod (6) is supported and fixed through the first auxiliary rod (7) and the second auxiliary rod (8), when the ground environment is uneven, the extension lengths of the first auxiliary rod (7) and the second auxiliary rod (8) are respectively adjusted to be respectively positioned on the same plane with the corresponding support rods (6), one side of the first mounting frame (10) is rotated through a rotating wheel, when the length of the first auxiliary rod (7) or the second auxiliary rod (8) is determined to be proper, the connecting rod (4) slides in the slot (20), thereby enabling the fixed rod (14) to slide on the surface of the movable rod (12), the second mounting rack (11) is moved, after the second mounting rack is moved, the first comb-shaped clamping groove (15) and the second comb-shaped clamping groove (16) are mutually clamped, so that the comb-shaped clamping grooves are stably connected, the service life of the original supporting rod (6) is conveniently prolonged, meanwhile, the first auxiliary rod (7) and the second auxiliary rod (8) can be conveniently adjusted, when the carrier loader is in an unstable section, the framework of the carrier loader is stable by matching with the support rod (6), meanwhile, compared with the supporting rods (6) with a plurality of supporting structures, the novel multifunctional support is smaller in size, more convenient to carry and more suitable for emergency tasks.

Example 2

The surface mounting of first cover splice block (2) has first auxiliary rod (7), and the surface mounting of second cover splice block has second auxiliary rod (8), and dwang (9) are all installed on the surface of the inside bracing piece (6) of first cover splice block (2) and second cover splice block (3), and first mounting bracket (10) are installed at one side top of dwang (9), and dwang (9) one side mid-mounting has second comb shape draw-in groove (16).

The leveling system comprises a first camera (25), a second camera (26) and a positioning device, the leveling system shoots a picture through the first camera (25) and the second camera (26), key data of the picture are extracted, the leveling system works in the process, the image input ranges of the first camera (25) and the second camera (26) are circular rings with the radius of one meter formed by taking two support rods (6) as the centers of circles respectively, and the leveling system adopts an image data identification method based on a convolutional neural network.

The leveling system records a flat ground picture in advance, inputs the flat ground picture through an input layer, trains data, extracts key characteristic data, classifies a neural network, repeats the training data until the precision is qualified if the recognition precision is not good, records and captures an image which takes a support rod as a circle and has a radius, inputs the image, recognizes the data after inputting the image to the input layer, transmits the data to the trained convolutional layer, transmits the data to the trained neural network, outputs a recognition result, judges the flat ground and judges the flat ground through a positioning device by sound or light spots or the combination of the sound and the light spots.

By adopting the technical scheme: the vehicle can be helped to find a stable ground through the arranged binocular recognition system, good external conditions are provided for the vehicle in an effort, the first sleeving block (2) and the second sleeving block (3) are respectively sleeved on the surfaces of the two supporting rods (6) at the rear part of the vehicle body (1) to enable the vehicle to be stably installed, the binocular recognition system records images when the vehicle runs, the accurate positions of the images are judged, the leveling system finds a flat ground, the positioning device positions a proper place, the first auxiliary rod (7) and the second auxiliary rod (8) are placed, and the supporting rods (6) are reinforced;

example 3

As shown in fig. 2-6, an image data identification system and method based on neural network, including first set of splice block (2), second set of splice block (3) and connecting rod (4), install connecting rod (4) between first set of splice block (2) and second set of splice block (3), second set of splice block (3) and first set of splice block (2) are all installed at automobile body (1) rear, bracing piece (6) are all installed to the bottom both sides of automobile body (1), the surface at one side bracing piece (6) is installed in first set of splice block (2), the surface at opposite side bracing piece (6) is installed in second set of splice block (3), mount pad (5) is still installed in the front of connecting rod (4), first infrared camera (27) and second infrared camera (28) are installed respectively to the top of mount pad (5), the bottom of mount pad (5) is equipped with slide rail (23), the surface sliding connection of slide rail (23) has two mount blocks (24), first camera (25) and second camera (26) are installed respectively to the surface of two mount pads (24), first camera (25) and second camera (26) are in the mount pad (5) bottom of mount pad (5).

The leveling system inputs a flat ground picture in advance, inputs the flat ground picture through an input layer, trains data, extracts key characteristic data, classifies a neural network, repeats the training data until the precision is qualified if the recognition precision is not good, records and captures an image which takes a support rod as a circle and takes one meter as a radius when the leveling system is in actual use, inputs the image and recognizes the data after inputting the image to the input layer, transmits the data to the trained convolutional layer, transmits the data to the trained neural network, outputs a recognition result, judges and judges the flat ground through a positioning device by sound or light spots or the combination of the sound and the light spots.

A pressure deformation fitting program is installed in the binocular recognition system, the pressure change of the supporting rod (6) can be detected through the first camera (25) and the second camera (26), and the first infrared camera (27) and the second infrared camera (28) can record the goods operation of the pontoon bridge transport vehicle.

By adopting the technical scheme: through the binocular identification system who is equipped with, first infrared camera and the infrared deformation of making a video recording of second can be to bracing piece (6) are caught, utilize two cameras of first camera (25) and second camera (26) can make the distance of judgement more accurate, the boat bridge transport vechicle parks the back, carry out pressure curve's fitting to its pressure that receives, can detect the pressure that bracing piece (6) received and the loss of bracing piece (6) in parking many times through the contrast between the pressure, the loss of bracing piece (6) is discerned, install the top at mount pad (5) simultaneously first infrared camera (27) and second infrared camera (28), can take notes the goods of boat bridge transport vechicle loading and unloading.

Example 4

As shown in fig. 1, 7, 8, 9 and 10, an image data recognition system and method based on neural network includes the following steps:

the method comprises the following steps: data input, wherein the data input finishes the pretreatment of the data, processes the data into more obvious characteristic data after the pretreatment, finally forms identification or training data, then normalizes the images (the video is composed of a plurality of frames of images), enables the images to resist geometric deformation through normalization, finds out the invariance of the images, and thereby knows that the images are originally one or a series of images, and the image data in the project is the UNIT data of 0-255 so that the normalization is required to be converted to be between 0-1;

step two: the convolution calculation layer design, convolution layers are used for extracting features in an image, each convolution layer can be provided with a plurality of convolution kernels representing the convolution layer, an original image is divided into three channels of R, G and B in the project, one image is decomposed into an input layer of 9 x 6 x 3, and three convolutions are respectively created for the input layer to generate three convolution kernels;

step three: constructing an excitation function, and using a LeakyRelu function to perform nonlinear mapping so as to improve the learning efficiency;

step four: in the pooling layer design, because the images needing to be processed in the system are too large, the characteristic dimension reduction is needed, the number of image data and parameters is compressed, the fitting is reduced, and meanwhile, the fault tolerance of the model is improved, 2 x 2 is selected in the system, and the maximum pooling is adopted, namely the maximum value of 4 points is taken;

step five: the fully connected layer design, which uses the high-level features to classify the input images into different classes, according to the high-level features of the input images outputted by the convolutional layer and the pooling layer. Therefore, the image classification task to be executed by the system is obtained, and the value with the maximum possibility is output.

For unstructured data, a Convolutional Neural Network (CNN) is adopted for supervised learning to realize classification and identification of the unstructured data, and meanwhile, a feature engineering is adopted for data to highlight data features, and the framework is shown in fig. 7;

meanwhile, according to the recognition purpose, a recognition framework is constructed, as shown in fig. 7, CNN includes one input layer, 4 special structural units (TraConv 0, mlpConv1, mlpConv2, and MlpConv 3), and two fully connected layers fc1 and fc2.

The network structure parameters of the CNN model are listed in table 1. As shown in the table, the CNN model is input as an image analyzed from the captured navicular video, and the image is converted into a size of 3 × 224 × 224 and then input as a structural element TraConv 0. C, H, W in table 1 represent the number of channels, height and width of the image, respectively. TraConv0 retains the conventional convolution form, comprising in sequence a convolution layer, a pooling layer, and a Local Response Normalization (LRN) layer.

TABLE 1 CNN network architecture parameters

After TraConv0, 3 building blocks were created in order: mlpConv1, mlpConv2 and MlpConv3. The 3 units have the same structure, the difference is that the step size and the padding of the first convolutional layer in the MlpConv1 are different from those of the latter two, the MlpConv1 structural unit comprises a convolutional layer conv1 with a convolutional kernel size of 3 × 3, the latter two convolutional layers cccp1_1 and cccp1_2 with convolutional kernel sizes of 1 × 1 are followed, and finally the hidden layer result is output after being maximally pooled.

As shown in table 1, the output of the pool2 layer includes 96 feature maps with a size of 6 × 6, each feature map is converted into a one-dimensional vector and sequentially spliced into a 3456-dimensional vector, which is denoted as vector a. Similarly, the output of pool3 is spliced into an 1152-dimensional vector, which is denoted as vector B. And then the vector A and the vector B are spliced to form a one-dimensional vector C with the length of 4608, and the vector C is used as the input of the fc1 layer. The vector C is subjected to dimension reduction to 160 dimensions after passing through the fc1 layer, and the dimension reduced vector is the final characteristic representation of the vehicle state extracted from the sample. The vector B of pool3 has more global vehicle features than the vector a extracted from pool 2. Considering that the classification should pay more attention to equipment motion local feature extraction than motion features, the vector a extracted from pool2 in the spliced vector C has a greater weight. The fusion of the global features and the local features preserves the feature information of equipment as much as possible from different scales, thereby improving the feature expression capability of the network.

In an actual test site, the shot heavy boat transport vehicle and the shot bridge deck comprise abundant local characteristics such as hydraulic rods, roll-over frames, support frames and the like besides the vehicle body outline, so that when the vehicle characteristics are extracted, the local characteristics and the global characteristics of the vehicle need to be considered, as shown in fig. 8;

the method comprises the steps of extracting picture data of key frames from video data according to the video data, and marking manual data according to different pictures. 7500 data labels are marked to serve as training data for training of a machine learning model, and the training data is shown in FIG. 9;

the key action of the equipment is identified from the video through the method, and the video acquisition time is read, so that the equipment action completion time is obtained, as shown in fig. 10.

The image data classification and identification method based on the improved convolutional neural network is adopted, the final erection process of the mechanical bridge is obtained as 'bridge feet are put down', the process is consistent with that of the actual video, and the experimental result verifies the effectiveness of the image data classification and identification method based on the improved convolutional neural network.

The invention is to be noted that the invention is an image data recognition system and method based on neural network, when in use, the first sleeve joint block (2) and the second sleeve joint block (3) are respectively sleeved on the surfaces of two support rods (6) at the rear part of the vehicle body (1) to ensure that the installation is stable, the binocular recognition system records the image of the environment, the accurate position of the image is conveniently judged, the leveling system searches for flat ground, the positioning device positions the proper place, the first auxiliary rod (7) and the second auxiliary rod (8) are placed to reinforce the support rods (6), if the proper position is not found, the support rods (6) are supported and fixed through the first auxiliary rod (7) and the second auxiliary rod (8), when the ground environment is uneven, the extension lengths of the first auxiliary rod (7) and the second auxiliary rod (8) are respectively adjusted to be respectively positioned on a plane together with the corresponding support rod (6), one side of the first mounting frame (10) is rotated through a rotating wheel, after the length of the first auxiliary rod (7) or the second auxiliary rod (8) is determined to be proper, the connecting rod (4) slides in the slot (20), so that the fixed rod (14) slides on the surface of the movable rod (12), the second mounting frame (11) is moved, and after the movement, the first comb-shaped clamping groove (15) and the second comb-shaped clamping groove (16) are mutually clamped to be stably connected, when the pontoon bridge transport vechicle is parked, can produce the recoil to bracing piece (6), cause the damage to bracing piece (6), catch the deformation of bracing piece (6) respectively through two mesh identification system, simultaneously, can fit pressure transformation curve, discern the loss of bracing piece (6), simultaneously, first infrared camera (27) and second infrared camera (28) can be to the freight logistics situation taking notes in the pontoon bridge transport vechicle.

The foregoing shows and describes the general principles and broad features of the present invention and advantages thereof. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims

1. An image data recognition system based on a neural network, characterized in that: the connecting rod is installed between the first sleeving connecting block and the second sleeving connecting block, the second sleeving connecting block and the first sleeving connecting block are installed behind a vehicle body, supporting rods are installed on two sides of the bottom end of the vehicle body, the first sleeving connecting block is installed on the surface of the supporting rod on one side, the second sleeving connecting block is installed on the surface of the supporting rod on the other side, and an installation seat is further installed on the front face of the connecting rod; the first camera and the second camera are matched with each other to form a binocular recognition system, the first camera and the second camera respectively detect support rods on two sides of the vehicle body, and the first camera and the second camera are connected with a leveling system; the surface mounting of first cover joint piece has first auxiliary rod, the surface mounting of second cover joint piece has the second auxiliary rod, inside first cover joint piece and the second cover joint piece the dwang is all installed on the surface of bracing piece, first mounting bracket is installed at one side top of dwang, dwang one side mid-mounting has second comb shape draw-in groove.

2. The neural network-based image data recognition system of claim 1, wherein: the utility model discloses a fixed pole, including first mounting bracket, runner, semicircular arc mid-mounting, the surperficial cover of movable rod is equipped with the sliding block, the dead lever is installed to the bottom of sliding block, the connecting rod is installed to the other end of dead lever.

3. The neural network-based image data recognition system of claim 1, wherein: the other end of first auxiliary rod and second auxiliary rod all is equipped with the second mounting bracket, the cooperation piece is installed to the other end of second mounting bracket, the cuboid opening is seted up at the middle part of cooperation piece, the fluting has all been seted up to the both sides of cooperation piece, two flutings have been seted up respectively at the both ends of connecting rod, the movable pulley is installed at the both ends of connecting rod, the movable pulley is in the grooved outside of cooperation piece, first comb shape draw-in groove is installed to the bottom of second mounting bracket, first comb shape notch has been seted up on the surface of first comb shape draw-in groove, second comb shape notch has been seted up on the surface of second comb shape draw-in groove, the length of first comb shape draw-in groove is greater than the length of second comb shape draw-in groove, first comb shape notch and the second comb shape draw-in groove on first comb shape draw-in groove and its surface and the second comb shape draw-in groove and the second comb shape notch size of surface match each other.

4. The neural network-based image data recognition system of claim 1, wherein: the leveling system comprises a first camera, a second camera and a positioning device, the leveling system shoots pictures through the first camera and the second camera and extracts key data of the pictures, the leveling system works in the process, the ranges of input images of the first camera and the second camera are circular rings with the radius of one meter formed by taking two support rods as the circle centers respectively, and the leveling system adopts an image data identification method based on a convolutional neural network.

5. The neural network-based image data recognition system of claim 1, wherein: the leveling system is characterized in that a flat ground picture is input in advance, input is carried out through an input layer, data training is carried out, key characteristic data are extracted, neural network classification is carried out finally, repeated identification is carried out if the identification precision is not good until the precision is qualified, when the leveling system is in actual use, an image which is captured and takes a support rod as a circle and one meter as a radius is input and input, the data are identified after being input to the input layer, the leveling system transmits the data to a trained coiling layer, then the data are transmitted to the trained neural network, then an identification result is output, and flat ground is judged and judged through a positioning device by sound or light spots or the combination of the sound and the light spots.

6. The neural network-based image data recognition system of claim 1, wherein: binocular identification system internally mounted has pressure deformation fitting program, enough detects the pressure variation of bracing piece through first camera and second camera, first infrared camera and second infrared camera can carry out the record to the pontoon bridge operation condition.

7. The use method of the image data recognition system based on the neural network as claimed in any one of claims 1 to 6, wherein: the using method comprises the following steps:

a, step a: the first sleeving connection block and the second sleeving connection block are respectively sleeved on the surfaces of two supporting rods at the rear part of the vehicle body to enable the supporting rods to be stably installed, the binocular recognition system is used for recording images of the environment, the accurate positions of the images are conveniently judged, the leveling system is used for finding flat ground, the positioning device is used for positioning a proper place, the first auxiliary rod and the second auxiliary rod are placed, and the supporting rods are reinforced;

step c: when the pontoon bridge equipment loads and unloads, can produce the recoil to the bracing piece, cause the damage to the bracing piece, catch the deformation of bracing piece respectively through two mesh identification systems, simultaneously, can fit pressure transformation curve, discern the loss of bracing piece, simultaneously, first infrared camera and second infrared camera can equip the operation condition to the pontoon bridge and take notes.

8. The use of a neural network-based image data recognition system as claimed in any one of claims 1 to 6, wherein: the using method comprises the following steps:

step four: designing a pooling layer, selecting 2 x 2 in the system, and taking the maximum value of 4 points by adopting maximum pooling;