A method and system for capturing seabed images with self-coding

Abstract

The present invention relates to the field of data processing technology, and in particular to a self-encoded seabed image capture method and system, comprising the following steps: collecting seabed image data through a collector; performing image preprocessing on the collected seabed image data; performing image compression processing on the preprocessed seabed image; and transmitting the compressed image in an integrated manner: a receiver receives a signal sent by the collector and performs image decompression to obtain the original seabed image data. The present invention uses a conversion formula to convert the color space, making it easier to achieve compression of seabed image data, facilitating data transmission and processing, and improving the speed of data processing and transmission. By using a compression model to compress and binarize the collected seabed image data, the accuracy of the internal convolution operation is reduced to achieve optimization of the system load and resource usage. The digital signal carrying image information is transmitted through sound waves, which has strong noise resistance, eliminates the use of cables, and has the characteristics of low power consumption.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and system for capturing seabed images with self-coding. Background Art

[0002] It is estimated that more than 95 percent of the Earth's oceans have never been observed, which means that we have explored less of the Earth's oceans than the Moon or Mars. Providing power supply for long-term underwater video research activities is a serious challenge for ocean exploration. The present invention provides a design concept for a battery-free wireless underwater camera that does not require a power source and can operate underwater for years. It can be used to monitor marine pollution or the growth of various underwater organisms. This wireless underwater camera captures image data by irradiating the object with light from a photodiode, encoding it into bits and sending it one bit at a time to a surface receiver using underwater backscatter. The system uses this binary information to reconstruct and process the image. However, in the existing technology, images are transmitted wirelessly and in a bit-by-bit mode, which is slow and time-consuming. Summary of the Invention

[0003] The purpose of the present invention is to solve the defects mentioned in the above background technology by proposing a method for capturing seabed images with self-coding.

[0004] The technical solution adopted in the present invention is as follows:

[0005] A method for capturing seabed images with self-coding is provided, comprising the following steps:

[0006] S1.1: Collect seabed image data through the collector;

[0007] S1.2: Perform image preprocessing on the collected seabed image data;

[0008] S1.3: Perform image compression processing on the pre-processed seabed image data;

[0009] S1.4: Integrated transmission of compressed seafloor image data:

[0010] S1.5: The receiver receives the signal sent by the collector and decompresses the image to obtain the original seabed image data.

[0011] As a preferred technical solution of the present invention: the image preprocessing operation in S1.2 determines the color space of the collected RGB three-channel seabed color image data and converts it into YCbCr representation:

[0012]

[0013] Among them, Y represents the brightness component, C bRepresents the blue chromaticity component, C r Represents the red chromaticity component, R, G, and B represent the three primary color components of red, green, and blue respectively.

[0014] As a preferred technical solution of the present invention: in S1.3, the pre-processed seabed image is compressed by constructing a compression model.

[0015] As a preferred technical solution of the present invention, the compression model is to quantize the input seabed image component data, and the quantization function is:

[0016]

[0017] Among them, sign(x) is the quantization function.

[0018] As a preferred technical solution of the present invention: the quantized seabed image component data is then passed through a convolution layer to obtain binary data:

[0019] Use x to represent the component data of the preprocessed ocean image. The autoencoder uses x∈R n =X is the input, which is mapped to h∈R p =Z,

[0020] h=sign(x)(Wx+b)

[0021] Among them, h is called the latent variable, W represents the weight matrix, b is the bias vector, W and b are randomly initialized and iteratively updated during the propagation process. After the convolution process is completed, a numerical value is added before the data to represent different component data, using 1 to represent the brightness component Y and 0 to represent the blue chromaticity component C. b , use -1 to represent the red chromaticity component C r .

[0022] As a preferred technical solution of the present invention: in the convolution function, the weight matrix W is iteratively updated according to the quantization function sign(x) to obtain W′∈{1,-1}. The update formula of W′ is as follows:

[0023]

[0024] Among them, σ is the learning parameter and P is the loss function.

[0025] As a preferred technical solution of the present invention: in S1.5, the collector sends the seabed image data after the convolution layer to the receiver through wireless radio frequency communication, and the receiver then reconstructs the binary image.

[0026] As a preferred technical solution of the present invention: after the receiver receives the complete seabed image data signal, the weight matrix W is iteratively updated according to the inverse quantization function to achieve reconstruction of the compressed binary image.

[0027] As a preferred technical solution of the present invention: the inverse quantization function is as follows:

[0028]

[0029] Among them, signf(x) is the inverse quantization function.

[0030] Provided is a self-encoded seabed image capture system, including:

[0031] Image acquisition module: used to collect seabed image data;

[0032] Image preprocessing module: used to preprocess the collected seabed image data;

[0033] Image compression module: used to compress and transmit pre-processed seabed image data;

[0034] Image sending module: used to send compressed seabed image data;

[0035] Image receiving module: used to receive the transmitted compressed seabed image data;

[0036] Image decoding module: used to decode and restore the received compressed seabed image data.

[0037] Compared with the prior art, the self-encoded seabed image capture method and system provided by the present invention have the following beneficial effects:

[0038] Using the YCbCr mathematical conversion formula makes it easier to compress seafloor image data, facilitating data transmission and processing, and increasing data processing and transmission speed. The convolution layer performs a convolution operation on the entire binary image while preserving the original image's accuracy without increasing the computational effort. This compression process significantly reduces the amount of transmitted data and transmission time. By using a compression model to compress and binarize the collected seafloor image data, the accuracy of the internal convolution operation is reduced, optimizing system load and resource usage. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A flow chart of a method according to a preferred embodiment of the present invention;

[0040] Figure 2 This is a system block diagram in a preferred embodiment of the present invention.

[0041] The meanings of the various marks in the figure are: 100, image acquisition module; 200, image preprocessing module; 300, image compression module; 400, image sending module; 500, image receiving module; 600, image decoding module. DETAILED DESCRIPTION

[0042] It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of this embodiment can be combined with each other. The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0043] Reference Figure 1 The preferred embodiment of the present invention provides a method for capturing seabed images with self-coding, comprising the following steps:

[0044] S1.1: Collect seabed image data through the collector;

[0045] S1.2: Perform image preprocessing on the collected seabed image data;

[0046] S1.3: Perform image compression processing on the pre-processed seabed image;

[0047] S1.4: Transmit the compressed image in an integrated manner:

[0048] S1.5: The receiver receives the signal sent by the collector and decompresses the image to obtain the original seabed image data.

[0049] The image preprocessing operation in S1.2 determines the color space of the collected RGB three-channel seabed color image data and converts it into YCbCr representation:

[0050]

[0051] Among them, Y represents the brightness component, C b Represents the blue chromaticity component, C r Represents the red chromaticity component, R, G, and B represent the three primary color components of red, green, and blue respectively.

[0052] In S1.3, the pre-processed seabed image is compressed by constructing a compression model.

[0053] The compression model is to quantize the input seabed image component data, and the quantization function is:

[0054]

[0055] Among them, sign(x) is the quantization function.

[0056] The quantized seabed image component data is then passed through the convolution layer to obtain binary data:

[0057] Use x to represent the component data of the preprocessed ocean image. The autoencoder uses x∈R n =X is the input, which is mapped to h∈R p =Z,

[0058] h=sign(x)(Wx+b)

[0059] Among them, h is called the latent variable, W represents the weight matrix, b is the bias vector, W and b are randomly initialized and iteratively updated during the propagation process. After the convolution process is completed, a numerical value is added before the data to represent different component data, using 1 to represent the brightness component Y and 0 to represent the blue chromaticity component C. b , use -1 to represent the red chromaticity component C r .

[0060] In the convolution function, the weight matrix W is iteratively updated according to the quantization function sign(x), and W is obtained. ′ ∈{1,-1},W ′ The update formula is as follows:

[0061]

[0062] Among them, σ is the learning parameter and P is the loss function.

[0063] In the above S1.5, the collector sends the seabed image data after the convolution layer to the receiver via wireless radio frequency communication, and the receiver then reconstructs the binary image.

[0064] After the receiver receives the complete seabed image data signal, it iteratively updates the weight matrix W according to the inverse quantization function to reconstruct the compressed binary image.

[0065] The inverse quantization function is as follows:

[0066]

[0067] Among them, signf(x) is the inverse quantization function.

[0068] Provided is a self-encoded seabed image capture system, including:

[0069] Image acquisition module 100: used to collect seabed image data;

[0070] Image preprocessing module 200: used to preprocess the collected seabed image data;

[0071] Image compression module 300: used for compressing and transmitting the pre-processed seabed image data;

[0072] Image sending module 400: used to send compressed seabed image data;

[0073] Image receiving module 500: used to receive the transmitted compressed seabed image data;

[0074] Image decoding module 600: used to decode and restore the received compressed seabed image data.

[0075] In this embodiment, the image acquisition module 100 is placed on the seabed to collect RGB three-channel seabed color image data. The image preprocessing module 200 determines the color space of the collected RGB three-channel seabed color image data and converts it into YCbCr representation:

[0076]

[0077] Among them, Y represents the brightness component, C b Represents the blue chromaticity component, C r Represents the red chromaticity component, R, G, and B represent the three primary color components of red, green, and blue respectively.

[0078] Using the YCbCr mathematical conversion formula, it is easier to compress seabed image data, facilitate data transmission and processing, and improve the speed of data processing and transmission.

[0079] The image compression module 300 receives the seabed image data preprocessed by the image preprocessing module 200 and compresses the preprocessed image using a compression model.

[0080] The pre-processed seabed image component data is input through a parameter-learnable bias layer and then enters the quantization layer. The quantization layer quantizes the input seabed image component data. The quantization function is:

[0081]

[0082] Among them, sign(x) is the quantization function.

[0083] The pre-processed seabed image is binarized through the quantization function, which simplifies the original complex seabed image data and facilitates the subsequent data compression and output.

[0084] After the seabed image data is quantized, it passes through the convolution layer to obtain binary data:

[0085] Use x to represent the component data of the preprocessed ocean image. The autoencoder uses x∈R n =X is the input, which is mapped to h∈R p =Z,

[0086]

[0087] Among them, h is called the latent variable, W represents the weight matrix, W i Indicates the update of the i-th weight matrix, i is the number of weight matrix updates, n is the total number of updates during the forward transmission of the weight matrix, b is the bias vector, W and b are randomly initialized and iteratively updated during the propagation process. After the convolution process is completed, a numerical value is added before the data to represent different component data, using 1 for the brightness component Y and 0 for the blue chromaticity component C. b , use -1 to represent the red chromaticity component C r Among them, the weight matrix W is iteratively updated according to the quantization function sign(x), and W is obtained ′ ∈{1,-1},W ′ The update formula is as follows:

[0088]

[0089] Among them, σ is the learning parameter and P is the loss function.

[0090] The convolution layer performs convolution operations on the entire binary image and retains the accuracy of the original image without increasing the amount of calculation, greatly reducing the amount of transmission and shortening the transmission time.

[0091] The image sending module 400 sends the compressed binary data of the seabed image to the image receiving module 400 through acoustic wave communication. The image receiving module 400 receives the signal sent by the image sending module 400 and then reconstructs the binary image.

[0092] During the image reconstruction process, the weight matrix W is iteratively updated through the inverse quantization function containing effective gradients.

[0093]

[0094] Among them, signf(x) is the inverse quantization function.

[0095] Realize the reconstruction of compressed binary images.

[0096] By compressing and transmitting the seabed image data, performing binarization processing, and representing the characteristics of each component of the seabed image data, the accuracy of the original seabed data can be well preserved.

[0097] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

[0098] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A method for capturing seabed images with self-encoding, characterized by: The steps include: S1.1: Collect seabed image data through the collector; S1.2: Perform image preprocessing on the collected seabed image data; S1.3: Perform image compression processing on the pre-processed seabed image data; S1.4: Integrated transmission of compressed seafloor image data: S1.5: The receiver receives the signal from the collector and decompresses the image to obtain the original seabed image data; In said S1.3, the pre-processed seabed image is compressed by constructing a compression model; The compression model is to quantize the input seabed image component data, and the quantization function is: ； in, is the quantization function; The quantized seabed image component data is then passed through the convolution layer to obtain binary data: Use x to represent the component data of the preprocessed ocean image. The autoencoder uses As input, map it to , ； Among them, h is called the latent variable, represents the weight matrix, Indicates the update of the weight matrix for the i-th time, i is the number of updates of the weight matrix, and n is the total number of updates during the forward transmission of the weight matrix. is the bias vector, and The two are randomly initialized and iteratively updated during the propagation process. After the convolution process is completed, numerical values ​​are added before the data to represent different component data. Represents the brightness component ,use Represents the blue chromaticity component ,use Represents the red chromaticity component .

2. The method for capturing seabed images with self-encoding according to claim 1, characterized in that: The image preprocessing operation in S1.2 determines the color space of the collected RGB three-channel seabed color image data and converts it into YCbCr representation: ； in, represents the brightness component, represents the blue chromaticity component, Represents the red chromaticity component, R, G, and B represent the three primary color components of red, green, and blue respectively.

3. The method for capturing seabed images with self-encoding according to claim 2, characterized in that: In the convolution function, the weight matrix According to the quantization function Perform iterative updates to obtain , The update formula is as follows: ； in, To learn the parameters, is the loss function.

4. The method for capturing seabed images with self-encoding according to claim 1, characterized in that: In S1.5, the collector converts the seabed image data after the convolution layer into electroacoustic data through a piezoelectric transducer, and sends the data to the receiver in the form of sound waves in the seawater. The receiver converts the sound into electricity through the piezoelectric transducer, and then reconstructs the binary image.

5. The method for capturing seabed images with self-encoding according to claim 4, characterized in that: After the receiver receives the complete seabed image data signal, it quantizes the weight matrix according to the inverse quantization function. Perform iterative updates to achieve reconstruction of the compressed binary image.

6. The method for capturing seabed images with self-encoding according to claim 1, characterized in that: The inverse quantization function is as follows: ； in, is the inverse quantization function.

7. A self-encoded seabed image capture system, characterized by: include: Image acquisition module (100): used for collecting seabed image data; Image preprocessing module (200): used for preprocessing the collected seabed image data; Image compression module (300): used for compressing and transmitting pre-processed seabed image data; The pre-processed seabed image is compressed by constructing a compression model. The specific process is as follows: The compression model is to quantize the input seabed image component data, and the quantization function is: ； in, is the quantization function; The quantized seabed image component data is then passed through the convolution layer to obtain binary data: Use x to represent the component data of the preprocessed ocean image. The autoencoder uses As input, map it to , ； Among them, h is called the latent variable, represents the weight matrix, Indicates the update of the weight matrix for the i-th time, i is the number of updates of the weight matrix, and n is the total number of updates during the forward transmission of the weight matrix. is the bias vector, and The two are randomly initialized and iteratively updated during the propagation process. After the convolution process is completed, numerical values ​​are added before the data to represent different component data. Represents the brightness component ,use Represents the blue chromaticity component ,use Represents the red chromaticity component ; Image sending module (400): used for sending compressed seabed image data; Image receiving module (500): used for receiving transmitted compressed seabed image data; Image decoding module (600): used for decoding and recovering received compressed seabed image data.

Images