An underwater image enhancement method, system and terminal based on brain visual principle

Abstract

The application belongs to the technical field of information technology services, and discloses an underwater image enhancement method and system based on brain visual principle and a terminal. The method extracts image shallow features in combination with the human brain visual imaging principle. In order to reduce the feature loss of the original image during subsequent feature fusion, the RGB features of the original image are fused with the direction, color and brightness features, and then the primary features of 12 channels are obtained. Secondly, in combination with the top-down neural circuit idea in the human brain vision, a deep learning network is constructed to further enhance the detail features and fuzzy features of the underwater image, and to suppress the irrelevant non-information features to realize deep asymmetric feature extraction. The underwater image enhancement method based on brain visual principle has a high PSNR index, and the distortion between the image processed by the algorithm and the original image is small. The image processed by the algorithm has more suitable brightness, contrast, structural features and detail features, and is more in line with human visual perception.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of information technology services, and particularly relates to an underwater image enhancement method and system based on brain visual principle and a terminal. BACKGROUND

[0002] With the development of marine economy, marine development has attracted the attention of many countries. Marine development involves underwater topography detection, marine resource protection, marine biodiversity research and other related fields, including energy, biology, geography, etc. Underwater images and videos are particularly important as underwater information carriers. In practical applications, high-quality underwater images and videos can provide accurate underwater target observation data and regular underwater target distribution. How to obtain high-quality underwater images has been a problem that has plagued marine researchers for many years.

[0003] Due to the complexity of the imaging environment, camera blur, scattering and absorption of light by suspended particles, insufficient light in deep water, and other factors, underwater images often have visual degradation effects such as low contrast, low brightness, color deviation, difficulty in distinguishing details, and obvious noise, which limits the application of underwater images.

[0004] Currently, underwater image enhancement usually adopts traditional methods and deep learning-based methods. Traditional methods can be further divided into physical model-based methods and non-physical model-based methods.

[0005] Non-physical model methods do not rely on specific imaging models and improve the visual effect of images to some extent. However, since they do not consider the optical characteristics of underwater image formation, they often introduce color deviation and over-saturation and unsaturation problems. Physical model-based methods mainly model the degradation process of underwater images and estimate parameters based on the model. Such methods include dark channel prior, red channel prior, and minimum information prior. Physical model-based methods rely too much on the modeling of underwater imaging, which is usually based on certain prior assumptions and has limitations in the generalization of underwater images to different scene applications.

[0006] With the rapid development of artificial intelligence, deep learning has achieved excellent results in various visual tasks, and deep learning is widely used in underwater image enhancement. Chen et al. use underwater imaging model to clarify the fuzzy underwater image, and use the processed clear image and the corresponding fuzzy image to construct a pair of data set for training underwater image enhancement network. Fabbri et al. proposed a underwater image enhancement network based on Wasserstein GAN, but the method has many model parameters. Slam et al. based on generative adversarial network realizes fast enhancement of underwater image, but the generalization of the method still needs to be improved, resulting in large color difference between part of the enhanced image and the reference image. Chen et al. combine feature pyramid network with attention mechanism, which performs well in deblurring, improving color cast and other aspects, but the effect of improving image details still needs to be improved.

[0007] Through the above analysis, the problems and defects of the prior art are:

[0008] The existing underwater image enhancement method improves the underwater image quality in vision, but there are still problems such as detail loss, serious color cast, poor contrast, insufficient feature extraction ability and the like in the image restoration effect. The image processing algorithm based on deep learning network has obvious defects, such as low real accuracy of image training set, poor network processing ability, high experimental equipment requirement index and the like. SUMMARY

[0009] In view of the problems existing in the prior art, the present application provides an underwater image enhancement method, system and terminal based on human visual imaging principle.

[0010] The purpose of the present application is to solve the attenuation and degradation of underwater image caused by light, and then to cause the difficulty in feature extraction of underwater image, and the serious color cast. These underwater images that have lost part of the features have caused great trouble to the development of digital ocean field, and limited the transmission accuracy of underwater information. Based on such problems, the present application proposes an underwater image feature enhancement method and system based on primate visual imaging mechanism.

[0011] The present application is realized in this way, an underwater image feature enhancement method based on primate visual imaging mechanism, comprising:

[0012] Step one, adopt SolidWorks to design sealing device to waterproof, water-tight package of camera module, get waterproof camera of USB interface, can be used for shooting underwater image and display on PC end;

[0013] Step two, adopt enhancement algorithm to process underwater image, get preliminary enhanced underwater image, as the training set of the algorithm;

[0014] Step three, shallow visual feature extraction is performed on the underwater collected image to obtain color features, direction features, brightness features and other features of the underwater collected image, and the features are fused with the original image R, G and B features;

[0015] Step four, a deep learning network is constructed to process the primary features based on the shallow features to obtain deep visual features;

[0016] Step five, a jump deep supervision module is constructed, and the extracted deep visual features are introduced into the deep supervision module in stages to supervise and correct the results;

[0017] Step six, a pre-arranged training set is used to train the network to obtain a trained deep asymmetric feature extraction model;

[0018] Step seven, the underwater image to be measured collected by the waterproof camera is input into the trained deep asymmetric feature extraction model to complete the feature enhancement of the underwater image to be measured;

[0019] Step eight, a Python human-computer interaction interface is designed to realize the visualization of the enhancement effect.

[0020] Another object of the present application is to provide an underwater image feature enhancement system based on the primate brain visual principle, which comprises:

[0021] A waterproof camera module is used to collect underwater information and transmit it back to the PC end;

[0022] A human brain visual shallow extraction module is used to process and extract multi-dimensional features with high flexibility;

[0023] A down-sampling module is used to reduce the image size for subsequent feature extraction of the target;

[0024] An up-sampling module is used to enlarge the image to prevent the loss of small targets after multiple down-sampling, improve the gradient flow and receptive field of the image, and improve the feature extraction capability;

[0025] A deep asymmetric feature extraction module is used to process the primary features based on the shallow features to obtain high-dimensional information;

[0026] A Python human-computer interaction interface is used to visualize the enhanced image.

[0027] Another object of the present application is to provide an information data processing terminal for realizing the underwater image enhancement system based on the primate brain visual principle.

[0028] In combination with the technical solutions and the technical problems solved above, the technical solutions to be protected by the present application have the following advantages and positive effects:

[0029] Firstly, the underwater image enhancement method based on primate brain visual principle provided by the present application uses a deep learning network on the basis of shallow feature extraction, constructs a deep asymmetric feature extraction module, retains more high-dimensional information features on the basis of image primary features, and can well realize accurate enhancement of underwater images with features blurred and color seriously deviated.

[0030] Secondly, the brain visual theory of primates is often applied to the medical imaging field, such as human brain imaging mechanism and human brain attention mechanism. The present application redesigns an underwater image enhancement method based on brain vision by referring to the brain visual theory of primates, applies the idea in the medical imaging field to the computer vision task. In the process of constructing the deep learning network, the reference availability of the data set and the coordination of the network training often confuse relevant scholars. The present application fully considers the trade-off relationship between the network complexity and the network robustness, adjusts the network parameters for many times, can make the network adaptively adjust the training parameters along with the complexity of the data set, is beneficial to further improving the simplicity of the underwater image enhancement method based on deep learning, and further enhances the stability of the underwater image enhancement system based on the primate brain visual principle.

[0031] Thirdly, the underwater image enhancement method based on the primate brain visual principle provided by the present application is tested and compared on multiple public underwater image data sets, and the results show that the PSNR, SSIM and VIF indexes of the method realized by the present application lead other classic algorithms. In addition, the image processed by the present application retains more detail features and structural features, and on the basis of meeting the human eye visual system, provides an excellent solution for the loss of image information features in the process of digital marine image transmission. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments of the present application will be briefly introduced as follows. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0033] Figure 1 is a primate brain visual imaging mechanism schematic diagram provided by the embodiments of the present application;

[0034] Figure 2 is a shallow feature extraction module based on brain vision provided by the embodiments of the present application;

[0035] Figure 3is a deep asymmetric feature extraction network structure schematic diagram provided by an embodiment of the present application;

[0036] Figure 4 is a jump deep supervision module structure schematic diagram provided by an embodiment of the present application;

[0037] Figure 5 is a contrast diagram of underwater images processed by different methods provided by an embodiment of the present application, wherein (a) is an original image, (b) is CycleGAN, (c) is UGAN, (d) is DEEPWAVE, (e) is IBLA, (f) is GB, (g) is DCP, (h) is UDCP, and (i) is the method of the present application;

[0038] Figure 6 is an image processing effect schematic diagram provided by an embodiment of the present application, wherein the first row is an original image, and the second row is a processing result image of the present application;

[0039] Figure 7 is a contrast diagram of underwater images processed by different methods based on canny operator detection provided by an embodiment of the present application, wherein (a) is an original image, (b) is CycleGAN, (c) is UGAN, (d) is DEEPWAVE, (e) is FunnieGan, (f) is IBLA, (g) is GB, (h) is DCP, (i) is UDCP, and (j) is the method of the present application. DETAILED DESCRIPTION

[0040] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application.

[0041] In view of the problems in the prior art, the present application provides an underwater image enhancement method, system and terminal based on brain visual principle, which will be described in detail below with reference to the accompanying drawings.

[0042] The specific process of the underwater image enhancement system based on primate brain visual principle in the embodiment is as follows:

[0043] The brain of a primate has strong image processing capability, and its imaging mechanism is to first extract primary features of an image, and then process the primary features to obtain high-dimensional information, such as Figure 1The present application proposes a new underwater image enhancement method, which has strong ability in feature extraction. The method of the present application is aimed at the three characteristics of underwater images, i.e. loss of detail features, loss of structural features and serious color deviation. Through classical image algorithms, combined with the images collected by the waterproof camera module underwater, a labeled image is generated, and then an underwater image dataset suitable for the environment is established. The deep asymmetric feature extraction model trained has good underwater image processing capability, can correct the feature loss and serious color deviation of underwater images, and retain more details of underwater images. The deep learning network constructed by the method has strong robustness, strong data processing capability and simple implementation, and can be widely applied to image clarification in ice and snow, fog and haze days, and enhancement of ocean information transmission capability.

[0044] The purpose of the present application is achieved as follows:

[0045] The method first collects underwater environment photos through an underwater camera module, enhances the collected photos using classical algorithms, and establishes a labeled underwater image training set. Through training using the training set with early labels, the reliability of the subsequent network can be improved. A shallow feature extraction module based on the principle of brain vision is established to extract the shallow features of the test image, including 4 color features (R, G, B, Y), 4 direction features (15°, 60°, 105°, 150°), and 1 brightness feature. In order to reduce the loss of features of the original image during subsequent feature fusion, the RGB features of the original image are fused with the direction, color and brightness features, and then 12 channel primary features are obtained. This can make the network pay more attention to the blurred areas and high frequency areas of the image caused by light scattering, and realize the shallow feature extraction of the input image. The structure of the brain vision shallow extraction module is as shown in Figure 2 The module can process feature extraction in color deviation areas with higher flexibility.

[0046] In traditional recursive fusion networks, the feature receptive field of small down-sampling multiple is small, suitable for processing small targets, and the resolution information of small scale features (deep layer) is insufficient, not suitable for small targets. In YOLOv3, the understanding of multi-scale detection is that the feature map of 1 / 32 size (deep layer) has high down-sampling multiple, so it has a large receptive field, suitable for detecting large target objects, and the feature map of 1 / 8 has a small receptive field, so it is suitable for detecting small targets. For small targets, small scale features cannot provide the necessary resolution information, so it is still necessary to combine large scale features, which means that when constructing the feature network, multiple down-sampling operations are easy to cause the loss of small targets.

[0047] Due to the complexity of underwater environment, there are small targets such as shells, small organisms and seaweed, and there is no image enhancement algorithm suitable for all types of underwater images at present. Based on the shallow brain visual extraction module, a fusion strategy is adopted to design a deep asymmetric feature extraction module, which includes a feature collection module (cover n block, CNB), a down-sampling module SPPF, an up-sampling module Usample and a deep asymmetric feature extraction network, improves the gradient flow and receptive field of the image, and improves the feature extraction capability, and the deep asymmetric feature extraction network structure is as shown in Figure 3

[0048] Finally, a jump deep supervision module (NOR model) is introduced, which jumps the feature information of the previous layer network, respectively extracts and performs linear interpolation, and completes feature fusion with a 1x1 convolutional network, and then obtains an enhanced digital color image, realizes the deep feature extraction network and image enhancement algorithm for underwater images. The structure of the jump deep supervision module is as shown in Figure 4

[0049] The embodiment of the present application uses Python to design the upper computer interface, displays the enhanced underwater image and the processing sign-in original image on the human-computer interaction interface, so that the processing comparison result is more intuitive.

[0050] The specific implementation technical scheme of the embodiment of the present application is as follows:

[0051] Step one, using SolidWorks to design a sealed shell to waterproof, water-tight package the camera module, get a waterproof camera with USB interface, which can be used for shooting underwater images and displaying on PC;

[0052] Step two, using the enhancement algorithm to process the underwater image to get the preliminary enhanced underwater image as the training set of the algorithm;

[0053] Step three, shallow visual feature extraction is performed on the underwater collected image to obtain color features, direction features, brightness features and other features of the underwater collected image, and fuse them with the original image R, G and B features;

[0054] Step four, constructing a deep learning network to process the primary features based on the shallow features to obtain deep visual features;

[0055] Step five, constructing a jump deep supervision module, introducing the extracted deep visual features into the deep supervision module in stages, and supervising and correcting the results;

[0056] Step six, using the pre-arranged training set to train the network to obtain the trained deep asymmetric feature extraction model;

[0057] ​​Step seven, input the underwater image collected by the waterproof camera into the trained deep asymmetric feature extraction model to complete the feature enhancement of the underwater image to be tested;

[0058] Step eight, use Python to design a human-computer interaction interface to realize the visualization of the enhancement effect.

[0059] The specific signal and data processing process of the embodiment of the application is as follows:

[0060] 1) Design a waterproof camera: use software such as SolidWorks to design a sealed shell, waterproof, water-tight packaging of the camera module, and thus obtain a waterproof camera with a USB interface. The camera can be used to shoot images underwater and display them on a PC.

[0061] 2) Image enhancement: use enhancement algorithms to process underwater images to obtain preliminary enhanced underwater images. This enhancement process can improve the clarity, contrast and other visual effects of the image, providing better input for subsequent processing.

[0062] 3) Shallow visual feature extraction: shallow visual feature extraction is performed on the underwater collected image. This includes extracting color features, direction features, brightness features and other features of the image, and fusing them with the R, G and B features of the original image. These features can provide structural and detailed information of the image.

[0063] 4) Deep visual feature extraction: build a deep learning network, input the shallow features, process them, and obtain deep visual features. The deep learning network can learn higher-level image feature representations through multiple convolution and pooling operations.

[0064] 5) Jump deep supervision: build a jump deep supervision module, introduce the extracted deep visual features into the deep supervision module, and supervise and correct the results. This process can adjust the network parameters according to the pre-set objective function through the back propagation algorithm, improve the discriminability and expression ability of the features.

[0065] 6) Network training: use the pre-arranged training set to train the network, and through repeated iteration and optimization, the network can learn better representations of underwater image features. The goal of training is to obtain a deep asymmetric feature extraction model that can accurately extract and enhance the features of underwater images.

[0066] 7) Underwater image feature enhancement: input the underwater image collected by the waterproof camera into the trained deep asymmetric feature extraction model to complete the feature enhancement of the underwater image to be tested. This process can improve the clarity, detail restoration and noise suppression of the image.

[0067] 8) Visualization interface: design a human-computer interaction interface using programming languages such as Python to visualize the enhancement effect. Through the interface, the original image and the enhanced image can be displayed in real time

[0068] The user can intuitively observe and evaluate the enhancement effect and perform interactive operations.

[0069] The above is a detailed description of the processing process of signals and data in the embodiment of the present application, including camera design, image enhancement, feature extraction, network training, and visualization interface.

[0070] The application embodiment of the present application provides an information data processing terminal, which is used to realize the underwater image enhancement system based on the brain visual principle.

[0071] As preferred, the embodiment of the present application provides an underwater image enhancement system based on the brain visual principle, which comprises:

[0072] A waterproof camera module is used to collect underwater information and transmit it back to the PC end.

[0073] A human brain visual shallow extraction module is used to process the feature extraction of the color cast area with high flexibility.

[0074] A downsampling module is used to reduce the image size for subsequent feature extraction of the target.

[0075] An upsampling module is used to enlarge the image to prevent the loss of small targets after multiple downsampling, improve the gradient flow and receptive field of the image, and improve the feature extraction capability.

[0076] A deep asymmetric feature extraction module is used to deeply process the shallow primary visual features to improve the high-dimensional information processing capability.

[0077] A Python human-computer interaction interface is used to visualize the enhanced image.

[0078] In step one, the waterproof camera module used is RYS-800WAF, an Ethernet conversion interface is used, both USB and Ethernet interfaces are reserved, a SolidWorks is used to design an underwater sealed shell to realize the waterproof camera design; the waterproof camera designed in the present application supports underwater shooting at a depth of 20 meters and can be applied to underwater measurement and underwater exploration.

[0079] In order to better restore the visual effect of underwater images and retain more complete detail features, the previous underwater image enhancement method may not achieve good restoration effect using only a single loss function. The present application comprehensively considers the absolute error loss L1 and the structural similarity loss SSIM, designs a composite loss function suitable for underwater images, and performs training. The formula is as follows:

[0080] L ALL = k1L1+ k2L SSIM

[0081] Where k1 and k2 are weight coefficients, and in this chapter, k1 = 0.8 and k2 = 0.2. The L1 loss and SSIM loss are introduced in detail as follows.

[0082] The L1 loss calculation formula is as follows:

[0083]

[0084] Where x is the output image, y is the reference image, and N is the total number of three-channel pixels of the image.

[0085] The SSIM loss formula is:

[0086] SSIM(x, y) = [l(x, y) α × c(x, y) β × s(x, y) λ ]

[0087] Where l(x, y) is the brightness similarity, c(x, y) is the contrast similarity, and s(x, y) is the structural similarity.

[0088] The brightness similarity l(x, y) is defined as:

[0089]

[0090] Where μ x and μ y are the average values of x and y, respectively, and c1 is an adjustment constant.

[0091] The contrast similarity c(x, y) measures the contrast difference between x and y, and the formula is:

[0092]

[0093] Where σ x and σ y are the standard deviations of x and y, respectively, and c2 is an adjustment constant.

[0094] The structural similarity s(x, y) measures the structural difference between x and y, and its definition is:

[0095]

[0096] Where σ xy is the covariance of x and y, and c3 is a constant to avoid a denominator of 0.

[0097] In practical applications, in order to simplify the calculation, generally set alpha = beta = lambda = 1, c3 = c2 / 2.

[0098] The present application is realized based on the Pytorch deep learning framework, and the experimental reasoning speed is: FPS 35.

[0099] Experimental platform: windows10, python, pytorch1)8, RTX3090.

[0100] Experimental setup: Adam optimizer, learning rate 0.0001, 150 epochs, 6 batchsize, image size 256*256

[0101] Dataset: UIEB public dataset

[0102] Comparative algorithm: input, cyclegan, funiegan, ugan, deepwave, IBLA, GB, DCP, UDCP, A GAM, output

[0103] Comparative index: PSNR, VIFP, SSIM, UIQM

[0104] In order to verify the effectiveness of the underwater image enhancement method based on the brain vision principle provided by the embodiment of the present application in underwater image enhancement and feature extraction, the experimental results are compared with several classical underwater image enhancement algorithms, including: using the cycle-consistent adversarial network (Cycle GAN, 2020), the U-Net for domain-free image enhancement (U-GAN, 2021), the deep neural network based on wavelength attribute (DeepWave, 2021), the underwater image restoration based on image blur and light absorption (IBLA, 2017), the single underwater image restoration of blue-green channel defogging and red channel correction (GB, 2016), the single image haze removal using dark channel prior (DCP, 2011), and the transmission estimation in underwater single image (UDCP, 2013). Part of the experimental results are shown in Figure 5 . Figure 6 It is an image commonly used to evaluate the performance of underwater image color correction. The image set given by the present scheme covers different color distortion types, low contrast and blur degradation problems, and can reflect most underwater degradation scenes. In order to obtain higher quality underwater images, further reconstruction and enhancement are carried out on the images after color correction to improve the overall contrast and clarity of the images, and the performance of the algorithm is analyzed from the aspects of subjective visual effect comparison and objective quality evaluation. The test images come from the open source UIEB dataset.

[0105] For the problem of serious loss of detail features in underwater images, the pre- and post-processed images and other classic algorithm images are compared using the Canny operator to test the feature enhancement ability of the algorithm, and the effect is shown in Figure 7 .

[0106] For the problem of serious color cast in underwater images, a set of images commonly used to evaluate the color correction performance of underwater images is used to test the color cast processing ability of the algorithm, and the effect is shown in Figure 6 .

[0107] Objective quality evaluation:

[0108] Subjective visual effect evaluation will be affected by human subjective consciousness, and objective evaluation is also needed to further verify the image enhancement effect. Four classic objective evaluation indicators are used in this scheme, namely Peak Signal-to-Noise Ratio (PSNR), Structural Similarity (SSIM), Visual Information Fidelity (VIF), Naturalness Image Quality Evaluator (NIQE), and Underwater Image Quality Measure (UIQM).

[0109] PSNR measures the quality of an image by calculating the global size of the pixel error between the image to be evaluated and the reference image. The larger the PSNR value, the smaller the distortion between the image to be evaluated and the reference image, i.e. the better the image quality. SSIM is an objective evaluation standard for image quality that conforms to the characteristics of the human visual system, mainly considering three key features of the image to be evaluated: luminance, contrast, and structure. The larger the SSIM value, the smaller the gap between the output image and the distortion-free image, i.e. the better the image quality. VIF measures the quality of the image to be evaluated by calculating the mutual information between the image to be evaluated and the reference image, and has strong theoretical support in information fidelity, which expands the connection between the image and the human eye. The value range of VIF is [0, 1], and the larger the VIF, the better the image quality. UIQM is a no-reference underwater image quality evaluation index based on the human visual system, which uses color measurement index (UICM), sharpness measurement index (UISM), and contrast measurement index (UIConM) as evaluation basis, and expresses UIQM as a linear combination of the three. The larger the value, the better the color balance, sharpness, and contrast of the image.

[0110] As can be seen from Table 1:

[0111] 1) The PSNR index of the method proposed in this scheme is higher, indicating that the distortion between the processed image and the original image is smaller, and more details are retained;

[0112] 2) The algorithm proposed in this scheme has relatively high SSIM and VIF indexes, indicating that the processed image has obtained more suitable brightness, contrast and structural features, and is more consistent with human visual perception;

[0113] Table 1 Image quality evaluation index

[0114]

[0115] In view of the above analysis, the core innovation points of the technical solutions provided by the embodiments of the present application mainly include:

[0116] 1. Deep learning method based on brain visual principle: through shallow visual feature extraction and deep visual feature processing of underwater images, the brain visual processing process is simulated to realize effective enhancement of underwater images.

[0117] 2. Jump deep supervision module: deep visual features are introduced in stages to supervise and correct the results, improving the precision and stability of the enhancement effect.

[0118] 3. Waterproof camera design and human-computer interaction interface: use SolidWorks to design waterproof sealing devices to realize underwater image acquisition; design a human-computer interaction interface through Python to realize the visualization of the enhancement effect.

[0119] The following are two specific embodiments of the technical solutions provided by the embodiments of the present application:

[0120] Embodiment 1: Underwater archaeological image enhancement

[0121] In the field of underwater archaeology, image clarity is crucial for discovering and analyzing cultural relics. In this embodiment, the underwater image enhancement method is applied to the processing of underwater archaeological images. After collecting underwater archaeological site images, the trained deep asymmetric feature extraction model is used for enhancement processing, thereby improving the image clarity and helping archaeologists more accurately analyze and identify cultural relics.

[0122] Embodiment 2: Underwater environmental monitoring image enhancement

[0123] In the field of underwater environment monitoring, high-quality images help to accurately evaluate the underwater environment. In this embodiment, the underwater image enhancement method is applied to the processing of underwater environment monitoring images. After collecting underwater ecological environment images, the trained deep asymmetric feature extraction model is used for enhancement processing, thereby improving the image clarity and enabling environmental monitoring personnel to more accurately evaluate the status of the underwater ecological environment.

[0124] The two embodiments are only examples, and in actual applications, appropriate implementation schemes can be selected according to specific needs and technical levels.

[0125] It should be noted that the embodiments of the present application can be realized by hardware, software or a combination of software and hardware. The hardware part can be realized by special logic; the software part can be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or a specially designed hardware. Those skilled in the art can understand that the above-mentioned devices and methods can be realized by computer executable instructions and / or included in processor control code, such as provided on a carrier medium, such as a magnetic disk, CD or DVD-ROM, a programmable memory, such as a read-only memory (firmware), or a data carrier, such as an optical or electronic signal carrier. The device of the present application and its modules can be realized by hardware circuit, such as ultra-large scale integrated circuit or gate array, semiconductor, such as logic chip, transistor, etc., or programmable hardware device, such as field programmable gate array, programmable logic device, etc., can also be realized by software executed by various types of processors, or by a combination of the above-mentioned hardware circuit and software, such as firmware.

[0126] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any modification, equivalent replacement and improvement within the technical range disclosed by the present application, which is within the spirit and principle of the present application, should be covered within the protection scope of the present application.

Claims

1. An underwater image enhancement method based on the principle of brain vision, characterized in that, The application relates to an underwater image enhancement method based on a brain visual principle. The method simulates the human brain visual processing process through shallow visual feature extraction and deep visual feature processing of underwater images, and realizes effective enhancement of the underwater images. The method introduces deep visual features in stages, and supervises and corrects the results; a waterproof sealing device is designed by using SolidWorks, and underwater image acquisition is realized; A man-machine interactive interface is designed by using Python, and the enhancement effect is visualized. The underwater image enhancement method based on the brain visual principle specifically comprises the following steps. In step one, a sealing device is designed by using SolidWorks to waterproof and water-tightly package a camera module, and a waterproof camera with a USB interface is obtained, which can be used for shooting underwater images and displaying the images on a PC end; In step two, an enhancement algorithm is used to process the underwater images, and preliminary enhanced underwater images are obtained as a training set of the algorithm; In step three, shallow visual features of the underwater collected images are extracted, color features, direction features, brightness features and other features of the underwater collected images are obtained, and the features are fused with original image R, G and B features; In step four, a deep learning network is constructed to process the primary features based on the shallow features and obtain deep visual features; In step five, a jump deep supervision module is constructed, the extracted deep visual features are introduced into the deep supervision module in stages, and the results are supervised and corrected; In step six, the network is trained by using the pre-arranged training set, and a trained deep asymmetric feature extraction model is obtained; In step seven, the underwater images to be measured collected by the waterproof camera are input into the trained deep asymmetric feature extraction model, and feature enhancement of the underwater images to be measured is completed; In step eight, a man-machine interactive interface is designed by using Python, and the enhancement effect is visualized.

2. The method for underwater image enhancement based on the principle of brain vision of claim 1, wherein, In the underwater image enhancement method based on the brain visual principle, the network pays more attention to the fuzzy area caused by light scattering and the detail features of the high-frequency area of the image, and realizes shallow feature extraction of the input image.

3. The method for underwater image enhancement based on the principle of brain vision of claim 1, wherein, The imaging mechanism of the underwater image enhancement method based on the brain visual principle is that primary features of the image are preliminarily extracted, and high-dimensional information is obtained after the primary features are processed.

4. An underwater image enhancement system based on the principle of brain vision using the underwater image enhancement method according to any one of claims 1 to 3, characterized in that, The application relates to an underwater image enhancement method based on a brain visual principle. The waterproof camera module is used for collecting underwater information and transmitting the information back to a PC end. The brain visual shallow extraction module is used for processing feature extraction of a color deviation area with high flexibility. The downsampling module is used for reducing the image, facilitating subsequent feature extraction of a target. The upsampling module is used for enlarging the image, preventing loss of small targets after multiple downsampling, improving gradient flow and a receptive field of the image, and improving feature extraction capacity. The deep asymmetric feature extraction module is used for deep processing of shallow primary visual features, and improving high-dimensional information processing capacity. The Python man-machine interactive interface is used for visualizing the enhanced image.

5. A computer device, the computer device comprising a memory and a processor, the memory storing a computer program, and the computer program being executed by the processor to enable the processor to execute the steps of the underwater image enhancement method based on the brain visual principle according to any one of claims 1-3.

6. A computer readable storage medium storing a computer program, the computer program, when executed by a processor, causing the processor to perform the steps of the method for underwater image enhancement based on the principle of human visual system according to any one of claims 1-3.

7. An information data processing terminal, the information data processing terminal being configured to implement the system for underwater image enhancement based on the principle of human visual system according to claim 4.

Images