Underwater code identification method and device

Abstract

The invention belongs to the technical field of nuclear power, and particularly relates to an underwater code recognition method and device. According to the invention, optimization is carried out on the basis of the Retina network, and the image is trained and output in an end-to-end manner; the generalization performance of the trained network is high, and a good effect can be achieved after any environment training; other prior knowledge is not needed; and the relation between local and overall images can be better processed. Therefore, the problems that when thermal disturbance and water turbulence disturbance exist in a nuclear radiation high-temperature and high-noise environment and a radioactive underwater environment, an imaging light field is non-uniform, an underwater image is distorted, defocused and the like, and equipment codes cannot be accurately recognized are effectively solved.

Description

Technical Field

[0001] The invention belongs to the field of nuclear power technology, and specifically relates to a method and a device for underwater code recognition. Background Art

[0002] Code recognition under underwater thermal disturbance and noise interference is a challenging task. When there is a heat source in the imaging light path, the target itself is a heat source, or there are other reasons in the water body that cause turbulent disturbances, the imaging light field becomes non-uniform, causing the underwater image to further twist, distort, and defocus. This will cause the code to lose details after being affected by noise, blur, distortion, and color correction, which seriously interferes with the recognition process. These factors deform the appearance of the code, resulting in a decrease in the accuracy of the recognition algorithm. Using image post-processing methods to restore degraded underwater images, restore image details, and improve image quality has always been a research hotspot and difficult problem in the field of underwater machine vision.

[0003] In recent years, the code recognition technology under underwater thermal disturbance and noise interference mainly uses image algorithms such as binarization, noise filtering, correlation domain analysis and AdaBoost to obtain results through three stages of image preparation, text recognition and post-processing. It has the disadvantages of many steps, low accuracy and much manual intervention. The mainstream deep learning methods include CTC, Attention and ACE frameworks. For underwater code recognition, these three methods have complex structures, large memory consumption, long inference time, and cannot be deployed on low-power micro devices.

[0004] In response to the above problems, it is urgent to reduce the complexity and resource overhead of recognition while ensuring recognition accuracy. Summary of the invention

[0005] In order to overcome the problems existing in the related art, a method and device for underwater code recognition are provided.

[0006] According to one aspect of an embodiment of the present disclosure, a method for underwater code recognition is provided, the method comprising:

[0007] The Undarkchannel pyramid is added to the backbone network P3, P4, P5 of the Retina network and the middle layer C3, C4, C5 of the image feature pyramid respectively, and the Retina backbone network and the Undarkchannel pyramid are fused to form a feature pyramid;

[0008] Use the feature pyramid to identify the target object and output the recognized code.

[0009] In one possible implementation, the darkchannel is determined using the following formula:

[0010] darkchannel=∑I min(r，g，b)

[0011] Among them, I min(r，g，b) is the minimum gray value of R, G, B in the image pixel;

[0012] And then use the following formula to determine the Undarkchannel:

[0013]

[0014] in, For I min(r，g，b) Perform normalization.

[0015] In one possible implementation, the bottom layer D3 of the Undarkchannel pyramid is determined by the output image using the following formula, and the size of D3 is consistent with P3 and C3;

[0016]

[0017] in for Undarkchannel;

[0018] The D4 of the Undarkchannel pyramid is obtained by downsampling D3, and the size of D4 is consistent with P4 and C4;

[0019] The D5 of the Undarkchannel pyramid is obtained by downsampling D4, and the size of D5 is consistent with P5 and C5.

[0020] In one possible implementation, P3 in the original Retina network is obtained by downsampling P2 and adding it to D3, as shown in the following formula:

[0021] P3=add(downsampling(P2),D3)

[0022] P4 in the original Retina network is obtained by downsampling P3 and adding it to D4, as shown in the following formula:

[0023] P4=add(downsampling(P3),D4)

[0024] P5 in the original Retina network is obtained by downsampling P4 and adding it to D5, as shown in the following formula:

[0025] P5=add(downsampling(P4),D5)

[0026] The add operation is a pixel addition operation, and downsampling is downsampling;

[0027] C5 is obtained by weighted operation of P5 and D5, as shown in the following formula:

[0028] C5=∑(P5 (n , i , j) *D5 (i , j) )

[0029] C4 is obtained by weighting P4 and D4 and fusing them with C5 after upsampling, as shown in the following formula:

[0030] C4=add(upsampling(C5),∑(P4 (n，i，j) *D4 (i，j) ))

[0031] C3 is obtained by weighting P3 and D3 and fusing them with C4 after upsampling, as shown in the following formula:

[0032] C3=add(upsampling(C4),∑(P3 (n，i，j) *D3 (i，j) ))

[0033] Among them, n is the dimension, i and j are the positions of feature points, upsampling is the upsampling operation, and add is the fusion operation.

[0034] In a possible implementation, when the encoding ratio on the target object is uniform, three anchors with a width and height of n:m are set.

[0035] In a possible implementation, when the target object code consists of letters and numbers, the categories during training are 26 letters and 10 numbers.

[0036] In a possible implementation, the coding sequence is outputted through position measurement according to the target coding arrangement sequence.

[0037] In a possible implementation, the position of each detected code is calculated, the Y-axis coordinate is calculated first, and then the X-axis coordinate is calculated, using the following formula, and the output order is prioritized according to the size of Y and X:

[0038] codes=output(set(I min (Y, X) ))

[0039] Among them, the encoding string is Y in the detected encoding set set, and the one with smaller X is output first. After output, the encoding is deleted and output in sequence until the encoding set set is empty.

[0040] In one possible implementation, multiple frames of the same coded target are taken from different angles and fields of view and input into the network together. The accuracy of the output result is estimated to obtain the codes of the multiple frames. The coded string is statistically analyzed and the maximum probability result is output using the following formula:

[0041]

[0042] Among them, result is the final output encoding string result, frame is the number of frames shot, codes is the output result of each frame, and the string output with the highest probability will be calculated.

[0043] According to another aspect of an embodiment of the present disclosure, a device for underwater code recognition is provided, the device comprising:

[0044] The network fusion module is used to add the Undarkchannel pyramid to the backbone network P3, P4, P5 of the Retina network and the middle layer C3, C4, C5 of the image feature pyramid, and fuse the Retina backbone network and the Undarkchannel pyramid to form a feature pyramid;

[0045] The recognition module is used to identify the target object using the feature pyramid and output the recognized code.

[0046] According to another aspect of an embodiment of the present disclosure, a device for underwater code recognition is provided, the device comprising:

[0047] processor;

[0048] a memory for storing processor-executable instructions;

[0049] Wherein, the processor is configured to execute the above method.

[0050] According to another aspect of an embodiment of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, and the computer program instructions implement the above method when executed by a processor.

[0051] The beneficial effects of the present disclosure are: the present disclosure chooses to optimize on the basis of the Retina network, and trains and outputs images end-to-end; the trained network has high generalization performance, and can have good results after training in any environment; no other prior knowledge is required; and the relationship between the local and the whole image can be better handled. This effectively solves the problem that when the monitoring equipment is in a high-temperature and high-noise environment of nuclear radiation, or in a radioactive underwater environment with thermal disturbances and turbulent water disturbances, the imaging light field becomes non-uniform, causing the underwater image to be distorted, defocused, etc., and the device code cannot be accurately identified. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 The figure is a flow chart of a method for underwater code recognition according to an exemplary embodiment.

[0053] Figure 2 The figure is a schematic diagram showing how to determine an Undark channel according to an exemplary embodiment.

[0054] Figure 3 It is a schematic diagram showing the fusion of the Retina network and the Undarkchannel pyramid according to an exemplary embodiment.

[0055] Figure 4 This is a schematic diagram of the coding arrangement in an example.

[0056] Figure 5 This is a schematic diagram of the output sequence of the encoding arrangement in an example.

[0057] Figure 6 The present invention is a block diagram of a device for underwater code recognition according to an exemplary embodiment. DETAILED DESCRIPTION

[0058] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0059] In the related art, since underwater targets are generally lighter in color and have large noise and thermal disturbances, to address this problem, the present disclosure proposes an Undarkchannel method to optimize the network and amplify the grayscale value and weight of light-colored targets.

[0060] Figure 1 The flowchart of a method for underwater code recognition according to an exemplary embodiment is shown. The method can be executed by a terminal device, wherein the terminal device can be a computer, for example, a server, a desktop computer, a laptop computer, etc. The terminal device can also be, for example, a user device, a vehicle-mounted device or a wearable device, etc. The embodiment of the present disclosure does not limit the type of the terminal device. Figure 1 As shown, the method includes:

[0061] Step 1: Add the Undarkchannel pyramid to the backbone network P3, P4, P5 of the Retina network and the middle layers C3, C4, C5 of the image feature pyramid, where D3, D4, D5 in the Undarkchannel pyramid are obtained by downsampling the corresponding layers. Through pixel addition and downsampling operations, the Retina backbone network and the Undarkchannel pyramid are fused to form a feature pyramid, further improving the network's coding detection of distortion and appearance deformation.

[0062] Step 2: Use the feature pyramid to identify the target object and output the identified code.

[0063] The present disclosure integrates the Retina network with the Undarkchannel pyramid to enhance the attention to the coding details and font distortion, so that the distorted coding under thermal disturbance is strengthened and trained, making it easier to judge the correct coding class.

[0064] As an example of this embodiment, Figure 2 As shown, the darkchannel is determined by the following formula:

[0065] darkchannel=∑I min(r，g，b)

[0066] Among them, I min(r，g，b) It is the minimum gray value of R, G, B in the image pixel.

[0067] And then use the following formula to determine the Undarkchannel:

[0068]

[0069] in, For I min(r，g，b) Perform normalization.

[0070] like Figure 3 As shown in the figure, the Undarkchannel pyramid is added between the backbone network P3, P4, P5 of the Retina network and the image feature pyramid C3, C4, C5. Among them, the bottom layer D3 of the Undarkchannel pyramid is determined by the output image using the following formula, and the size of D3 is consistent with P3 and C3.

[0071]

[0072] in For Undarkchannel.

[0073] The D4 of the Undarkchannel pyramid is obtained by downsampling D3, and the size of D4 is consistent with P4 and C4.

[0074] The D5 of the Undarkchannel pyramid is obtained by downsampling D4, and the size of D5 is consistent with P5 and C5. The P3 in the original Retina network is obtained by downsampling P2 and adding it to D3, as shown in the following formula:

[0075] P3=add(downsampling(P2),D3)

[0076] P4 in the original Retina network is obtained by downsampling P3 and adding it to D4, as shown in the following formula:

[0077] P4=add(downsampling(P3),D4)

[0078] P5 in the original Retina network is obtained by downsampling P4 and adding it to D5, as shown in the following formula:

[0079] P5=add(downsampling(P4),D5)

[0080] The add operation is a pixel addition operation, and downsampling is downsampling.

[0081] C5 is obtained by weighted operation of P5 and D5, as shown in the following formula:

[0082] C5=∑(P5 (n , i , j) *D5 (i , j) )

[0083] C4 is obtained by weighting P4 and D4 and fusing them with C5 after upsampling, as shown in the following formula:

[0084] C4=add(upsampling(C5),∑(P4 (n，i，j) *D4 (i，j) ))

[0085] C3 is obtained by weighting P3 and D3 and fusing them with C4 after upsampling, as shown in the following formula:

[0086] C3=add(upsampling(C4),∑(P3 (n，i，j) *D3 (i，j) ))

[0087] Among them, n is the dimension, i and j are the positions of feature points, upsampling is the upsampling operation, and add is the fusion operation.

[0088] In this way, the present disclosure fuses the original Retina network with the Undarkchannel pyramid, increases the network's attention to coding details and font distortion, and allows distorted coding under thermal disturbances to be trained more intensively, making it easier to determine the correct coding class. The feature pyramid in the original Retina network is formed by fusing the backbone network with the Undarkchannel pyramid and then performing a weighted operation on it. This further improves the network's ability to detect distorted and deformed coding.

[0089] In one possible implementation, since the anchors in the original Retina network are composed of 9 different sizes, when the encoding ratio on the target object is unified, it is changed to 3 anchors with a width and height ratio of n:m. Reducing the number of anchors can improve the detection speed.

[0090] In a possible implementation, when the target object code consists of digits and letters, the number of classes during training is 36, namely, 26 letters and 10 digits.

[0091] like Figure 4 and Figure 5 As shown, the target codes can be arranged in order and the code sequence can be output through position measurement, which is convenient for subsequent system to query the corresponding code.

[0092] The position of each detected code is calculated, the Y axis coordinate is calculated first, and then the X axis coordinate is calculated. The output order is prioritized according to the size of Y and X. The formula is as follows:

[0093] codes=output(set(I min (Y, X) ))

[0094] The encoding string is the Y in the detected encoding set, and the one with the smaller X is output first, and then deleted. The encoding is output in sequence until the encoding set is empty.

[0095] Due to the large influence of underwater image noise, uneven noise distribution, and font distortion caused by thermal disturbance, it is extremely easy to affect the accuracy of code recognition. This solution takes multiple frames of the same coded target through different fields of view and angles, inputs them into the network together, estimates the accuracy of the output results, and obtains the codes of multiple frames. The coded string is statistically analyzed, and the following formula is used to output the result with the maximum probability:

[0096]

[0097] Among them, result is the final output encoding string result, frame is the number of frames shot, and codes is the output result of each frame. The string with the highest probability will be calculated and output.

[0098] In this way, the present invention aims to solve the problem of large noise impact on underwater images by shooting multiple frames with different fields of view and angles, inputting the codes of multiple frames into the network, and estimating the accuracy of the output results, and finally outputting the code string result with the highest probability.

[0099] According to another aspect of an embodiment of the present disclosure, a device for underwater code recognition is provided, the device comprising:

[0100] The network fusion module is used to add the Undarkchannel pyramid to the backbone network P3, P4, P5 of the Retina network and the middle layer C3, C4, C5 of the image feature pyramid, and fuse the Retina backbone network and the Undarkchannel pyramid to form a feature pyramid;

[0101] The recognition module is used to identify the target object using the feature pyramid and output the recognized code.

[0102] For the description of the above-mentioned device, please refer to the description of the above-mentioned method, which will not be repeated here.

[0103] Figure 6 1900 is a block diagram of an underwater code recognition device according to an exemplary embodiment. For example, the device 1900 can be provided as a server. Figure 6 , the apparatus 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932 for storing instructions, such as an application, that can be executed by the processing component 1922. The application stored in the memory 1932 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute instructions to perform the above method.

[0104] The device 1900 may also include a power supply component 1926 configured to perform power management of the device 1900, a wired or wireless network interface 1950 configured to connect the device 1900 to a network, and an input / output (I / O) interface 1958. The device 1900 may operate based on an operating system stored in the memory 1932, such as Windows Server™, MacOS X™, Unix™, Linux™, FreeBSD™, or the like.

[0105] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions, which can be executed by the processing component 1922 of the device 1900 to perform the above method.

[0106] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0107] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0108] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0109] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0110] Various aspects of the present disclosure are described herein with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0111] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0112] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0113] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0114] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for underwater code recognition, characterized in that: The method comprises: The Undarkchannel pyramid is added to the backbone network P3, P4, P5 of the Retina network and the middle layer C3, C4, C5 of the image feature pyramid respectively, and the Retina backbone network and the Undarkchannel pyramid are fused to form a feature pyramid; Use the feature pyramid to identify the target object and output the recognized code.

2. The method according to claim 1, characterized in that The darkchannel is determined using the following formula: darkchannel=∑I min(r，g，b) Among them, I min(r，g，b) is the minimum gray value of R, G, B in the image pixel; And then use the following formula to determine the Undarkchannel: in, For I min(r，g，b) Perform normalization.

3. The method according to claim 1, characterized in that The bottom layer D3 of the Undarkchannel pyramid is determined by the output image using the following formula, and the size of D3 is consistent with P3 and C3; in for Undarkchannel; The D4 of the Undarkchannel pyramid is obtained by downsampling D3, and the size of D4 is consistent with P4 and C4; The D5 of the Undarkchannel pyramid is obtained by downsampling D4, and the size of D5 is consistent with P5 and C5.

4. The method according to claim 1, characterized in that P3 in the original Retina network is obtained by downsampling P2 and adding it to D3, as shown in the following formula: P3=add(downsampling(P2),D3) P4 in the original Retina network is obtained by downsampling P3 and adding it to D4, as shown in the following formula: P4=add(downsampling(P3),D4) P5 in the original Retina network is obtained by downsampling P4 and adding it to D5, as shown in the following formula: P5=add(downsampling(P4),D5) The add operation is a pixel addition operation, and downsampling is downsampling; C5 is obtained by weighted operation of P5 and D5, as shown in the following formula: C5=∑(P5 (n，i，j) *D5 (i，j) ) C4 is obtained by weighting P4 and D4 and fusing them with C5 after upsampling, as shown in the following formula: C4=add(upsampling(C5),∑(P4 (n，i，j) *D4 (i，j) )) C3 is obtained by weighting P3 and D3 and fusing them with C4 after upsampling, as shown in the following formula: C3=add(upsampling(C4),∑(P3 (n，i，j) *D3 (i，j) )) Among them, n is the dimension, i and j are the positions of feature points, upsampling is the upsampling operation, and add is the fusion operation.

5. The method according to claim 1, characterized in that When the encoding ratio on the target object is uniform, three anchors with width and height of n:m are set.

6. The method according to claim 1, characterized in that When the target object code consists of alphanumeric characters, the categories during training are 26 letters and 10 digits.

7. The method according to claim 1, characterized in that According to the target coding sequence, the coding sequence is output through position measurement.

8. The method according to claim 1, characterized in that The position of each detected code is calculated, the Y-axis coordinate is calculated first, and then the X-axis coordinate is calculated. The output order is prioritized according to the size of Y and X using the following formula: codes=output(set(I min(Y，X) )) Among them, the encoding string is Y in the detected encoding set, and the one with smaller X is output first. After output, the encoding is deleted and output in sequence until the encoding set is empty.

9. The method according to claim 1, characterized in that: The same coded target is photographed in multiple frames at different angles and in different fields of view, and the frames are input into the network together. The accuracy of the output result is estimated to obtain the codes of multiple frames. The coded string is statistically analyzed by probability, and the maximum probability result is output using the following formula: Among them, result is the final output encoding string result, frame is the number of frames shot, codes is the output result of each frame, and the string output with the highest probability will be calculated.

10. A device for underwater code recognition, characterized in that: The device comprises: The network fusion module is used to add the Undarkchannel pyramid to the backbone network P3, P4, P5 of the Retina network and the middle layer C3, C4, C5 of the image feature pyramid, and fuse the Retina backbone network and the Undarkchannel pyramid to form a feature pyramid; The recognition module is used to identify the target object using the feature pyramid and output the recognized code.

11. A device for underwater code recognition, characterized in that: The device comprises: processor; a memory for storing processor-executable instructions; The processor is configured to execute the method according to any one of claims 1 to 10.

12. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 10 is implemented.

Images