Image recognition system and method based on big data technology

Abstract

The invention relates to the technical field of marine image recognition, discloses an image recognition system and method based on a big data technology, and aims to solve many challenges faced by marine image target recognition and labeling in the fields of marine environment monitoring, marine traffic management, resource exploration and the like. In order to solve the problems that marine targets are diverse in form, prone to being interfered by natural factors, complex in background and the like, the method comprises the steps that firstly, images are collected through high-resolution equipment, and denoising, contrast enhancement and graying processing are conducted; and then, an improved DeepLabV3 + network is utilized to construct a recognition model, the model is fused into an adaptive feature fusion module, and a loss function is optimized, so that the recognition precision and robustness are improved. By means of the method, the uncertainty and complexity of the marine environment can be effectively overcome, accurate recognition and intelligent labeling of ships, buoys, marine organisms and other targets are achieved, and powerful support is provided for intelligent application in the marine field.

Description

Technical Field

[0001] The present invention relates to the field of marine image recognition technology, and in particular to an image recognition system and method based on big data technology. Background Art

[0002] Intelligent analysis and annotation technology of marine images plays a vital role in the fields of marine environment monitoring, marine traffic management, marine resource exploration, etc. However, due to the particularity and complexity of the marine environment, target recognition and annotation in marine images face a series of challenges, which greatly limit the effective application of related technologies.

[0003] Targets in marine images, such as ships, buoys, and marine life, often show a variety of shapes and sizes. Ships may have different appearances due to their type, purpose, and loading status; buoys may have uncertain positions and different shapes due to factors such as weather and currents; and marine life is even more diverse and has a variety of shapes. This diversity and uncertainty brings great difficulties to the extraction and recognition of target features.

[0004] To make matters more complicated, marine images are easily affected by natural factors such as waves and fog. The ups and downs of the waves can cause the position and shape of the target in the image to change, and even cause the target to be partially blocked or completely disappear; while the presence of fog can reduce the clarity and contrast of the image, making the target features less obvious and easily confused with other background elements.

[0005] In addition, the background of marine images is also complex and changeable. The randomness of sea ripples, the unpredictable sky, the obstruction of clouds, and the outline of distant mountains, these background elements not only occupy most of the image space, but also their texture, color, brightness and other features are often similar to or overlap with the target features, further increasing the difficulty of target recognition.

[0006] In summary, target recognition and annotation technology in marine images faces multiple challenges, such as diverse target morphology, susceptibility to natural factors, and complex and changeable background. These challenges require us to fully consider the particularity and complexity of the marine environment when designing intelligent annotation methods, and adopt effective technical means to enhance the ability to extract target features and improve the accuracy and robustness of target recognition. Summary of the invention

[0007] The purpose of the present invention is to provide an image recognition system and method based on big data technology to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solution: an image recognition method based on big data technology, the method comprising:

[0009] Step 1: Collecting marine image data: using high-resolution camera equipment to capture marine image data containing the target object;

[0010] Step 2: Preprocessing the marine image data: Preprocessing the captured marine image data, including image denoising, contrast enhancement and grayscale processing;

[0011] Step 3: Build a maritime target recognition model: Build a maritime target recognition model based on the improved DeepLabV3+ network architecture. The model is used to identify and classify specific targets in maritime images. The model has an embedded adaptive feature fusion module and optimizes the loss function in the original DeepLabV3+ network.

[0012] Step 4: Identify and label maritime targets: Input the maritime image preprocessed in step 2 into the trained maritime target recognition model, perform the target recognition task, and output the labeling results containing the target category and location information.

[0013] Preferably, the implementation steps of the adaptive feature fusion module include:

[0014] S1: The input features are processed at multiple scales through the ASPP module to generate feature maps of different scales;

[0015] S2: Apply the attention mechanism to the feature maps of each scale and assign different weights to features of different regions;

[0016] S3: After unifying all feature maps to the same size using upsampling technology, perform pixel-by-pixel weighted summation;

[0017] S4: The fused feature map is further refined through the convolutional layer to generate high-level features for final classification and positioning.

[0018] Preferably, the optimized loss function includes: introducing Smooth L1 loss to replace the original L2 loss as the bounding box regression loss, which is defined as:

[0019]

[0020] Where x is the difference between the predicted value and the true value. When |x|<1, the loss is 0.5x 2 ; When |x|≥1, the loss is |x|-0.5.

[0021] Preferably, the optimized loss function further includes: using a combination of Dice coefficient loss and cross entropy loss as classification loss, and the Dice coefficient loss is defined as:

[0022]

[0023] Among them, A is the predicted segmentation result, B is the true label, |·| represents the number of elements in the set, and by minimizing Dice Loss, a high overlap between the predicted results and the true labels is promoted; at the same time, the cross entropy loss between the predicted classification results and the true labels is calculated. The cross entropy loss is defined as:

[0024]

[0025] Where N represents the number of categories, y i represents the indicator value of the i-th category in the true label, p i Represents the probability of prediction as the i-th category;

[0026] The Dice coefficient loss and the cross entropy loss are weighted and summed to balance the two losses. The calculation formula is:

[0027] Total Loss=λ×Dice Loss+(1-λ)×Cross Entropy Loss

[0028] Among them, λ is a hyperparameter used to adjust the relative importance of each part of the loss.

[0029] Preferably, the image denoising adopts Gaussian filtering to eliminate the noise in the image, and the value of each pixel in the image is replaced by the weighted average of the pixel values ​​in its neighborhood. The calculation formula is:

[0030]

[0031] Where G(x,y) is the value of the Gaussian function at the two-dimensional coordinate (x,y) of the image, x and y are the coordinates in two-dimensional space, σ is the standard deviation of the Gaussian distribution, e is the base of the natural logarithm, and π is the circumference of a circle.

[0032] Preferably, the step of preprocessing the contrast enhancement in the marine image data specifically includes:

[0033] Calculate the brightness histogram of the input marine image and count the number of pixels at each gray level;

[0034] According to the histogram distribution, determine the minimum value I of the image brightness min and the maximum value I max ;

[0035] Apply the linear stretching algorithm to map the original grayscale value I(x,y) to the new grayscale value I′(x,y);

[0036] The mapped image is traversed, and the grayscale value of each pixel is updated to I′(x, y) to obtain a contrast-enhanced marine image.

[0037] Preferably, the method for graying the marine image data is:

[0038] a. Select the weighted method to convert the color image into a grayscale image, and use the formula GRAY = 0.3*R + 0.59*G + 0.11*B to calculate the grayscale value. R, G, and B in the formula represent the components of the pixel in the red, green, and blue color domains respectively;

[0039] b. Read the image data of the sea image, traverse each pixel in the image, and obtain the RGB color value corresponding to each pixel;

[0040] c. Apply the color value conversion method, call the formula in step a to convert the RGB color value of each pixel obtained in step b to obtain the corresponding grayscale value;

[0041] d. Apply the converted grayscale value to each pixel to generate a grayscale image of the sea image.

[0042] Preferably, an image recognition system based on big data technology comprises:

[0043] Image acquisition module: equipped with high-resolution camera equipment to capture marine image data containing target objects and transmit the acquired image data to subsequent processing modules;

[0044] Preprocessing module: receives the image data transmitted by the image acquisition module and performs image denoising, contrast enhancement and grayscale processing;

[0045] Model building and training module: Based on the improved DeepLabV3+ network architecture, a maritime target recognition model is built. The model has an embedded adaptive feature fusion module and optimizes the loss function to improve recognition accuracy. This module is also responsible for the model training process, including the preparation of data sets, adjustment of model parameters, and formulation of training strategies.

[0046] Target recognition and labeling module: receives the image data processed by the preprocessing module, uses the trained marine target recognition model to perform target recognition, and outputs labeling results containing target category and location information;

[0047] Control and interface module: responsible for the coordination and control between the various modules of the system, including the transmission of image data, the configuration of model parameters, the output of recognition results, and the interactive interface with the external system.

[0048] Preferably, an electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned image recognition method when executing the computer program.

[0049] Preferably, a storage medium stores a computer program, which implements the above-mentioned image recognition method when executed by a processor.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] This invention effectively addresses the problem of diverse shapes and sizes of marine targets by using high-resolution image acquisition equipment and advanced DeepLabV3+ network architecture, combined with an adaptive feature fusion module. This method can deeply explore target features, improve the model's recognition accuracy for various marine targets such as ships, buoys, and marine life, and accurately mark targets even in complex and changeable marine environments.

[0052] In view of the impact of natural factors such as waves and fog on the quality of marine images, this paper introduces contrast enhancement technology in the preprocessing stage, which effectively improves the clarity and contrast of the image, making the target features more prominent. At the same time, through the design of optimized loss functions, such as the combination of Smooth L1 loss and Dice coefficient loss, the model is further enhanced to be robust to changes in target position and morphology, reducing misidentification and missed recognition caused by natural factors.

[0053] In the face of the complex and changeable background of marine images, the present invention uses the ASPP module to extract multi-scale features and combines it with the attention mechanism to enable the model to automatically focus on key target features and effectively suppress background interference. This not only improves the accuracy of target recognition, but also greatly reduces the confusion between background elements and target features, and improves the accuracy and efficiency of annotation. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a step diagram of the image recognition method based on big data technology according to the present invention;

[0055] Figure 2 This is the implementation flow chart of the adaptive feature fusion module;

[0056] Figure 3 This is the flowchart of image grayscale processing. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 creative work are within the scope of protection of the present invention.

[0058] See also Figure 1-3The present invention provides a technical solution: an image recognition method based on big data technology, the steps of the method include:

[0059] Step 1: Collect marine image data

[0060] High-resolution video equipment, such as high-definition cameras or photography devices mounted on drones, is used to continuously monitor the marine environment and capture marine image data including target objects such as ships, buoys, and marine life. To ensure image quality, the video equipment must have high definition, high dynamic range, and good anti-shake performance to obtain clear and stable image data in a complex and changing marine environment.

[0061] Step 2: Preprocessing marine image data

[0062] The captured marine image data is preprocessed to eliminate noise, enhance contrast, and perform grayscale processing. First, algorithms such as Gaussian filtering or mean filtering are applied to image denoising to reduce noise interference introduced by natural factors such as waves and fog. Then, histogram equalization or contrast stretching technology is used to enhance the contrast of the image to make the target features more prominent. Finally, the color image is converted to a grayscale image to reduce the computational complexity of subsequent processing.

[0063] Step 3: Build a maritime target recognition model

[0064] A marine target recognition model is constructed based on the improved DeepLabV3+ network architecture. DeepLabV3+ is an advanced deep learning model that is particularly suitable for image segmentation and target detection tasks. In this implementation, the following improvements are made to the original DeepLabV3+ network:

[0065] Adaptive feature fusion module: Introduces an adaptive feature fusion mechanism to effectively fuse feature maps at different levels to improve the model's ability to recognize diverse shapes and sizes of maritime targets.

[0066] Loss function optimization: To address issues such as target occlusion and complex background in marine images, the original cross entropy loss function is improved by introducing loss functions such as focal loss or IoU loss that are more suitable for unbalanced datasets to improve the recognition accuracy and robustness of the model.

[0067] The constructed model is trained on a large marine image dataset. By iteratively optimizing the model parameters, it gradually learns to recognize and classify specific targets in marine images.

[0068] Step 4: Identify and mark maritime targets

[0069] The marine image preprocessed in step 2 is input into the trained marine target recognition model. The model analyzes the input image pixel by pixel, identifies the target object in the image, and outputs the annotation results including the target category (such as ships, buoys, marine life, etc.) and location information (such as the coordinates of the target in the image, bounding box, etc.). These annotation results can be further used for intelligent applications in the fields of marine environment monitoring, maritime traffic management, marine resource exploration, etc.

[0070] The present invention will be further described below in conjunction with Examples 1 to 4:

[0071] Embodiment 1:

[0072] The implementation of the adaptive feature fusion module of the present invention will be described in detail below with reference to specific examples.

[0073] In order to achieve accurate recognition and positioning of targets in marine images, this paper designs an adaptive feature fusion module, which can make full use of multi-scale features and enhance the expression of key features through the attention mechanism. The specific implementation steps are as follows:

[0074] S1: Multi-scale feature extraction

[0075] The ASPP (Atrous Spatial Pyramid Pooling) module is used to perform multi-scale processing on the input features. The ASPP module captures contextual information of different scales in the image by applying convolution kernels with different dilation rates in parallel. For example, four 3x3 convolution kernels with dilation rates of 1, 6, 12, and 18 are set to perform convolution operations on the input feature map to generate four sets of feature maps with different receptive fields. In this way, each set of feature maps can capture target features of different scales, such as small targets, medium targets, and large targets, providing rich multi-scale information for subsequent feature fusion.

[0076] S2: Attention Mechanism Weighting

[0077] An attention mechanism is applied to feature maps of each scale to assign different importance weights to features in different regions. Specifically, a channel attention mechanism is used to weight the channels of each feature map. First, the spatial dimension of each feature map is compressed into a single value through a global average pooling layer to obtain a channel descriptor. Then, two fully connected layers and a ReLU activation function are used to learn the dependencies between channels and generate channel weights. Finally, the channel weights are multiplied with the original feature map to achieve feature weighting, enhance the expression of key features, and suppress irrelevant features.

[0078] S3: Feature map unification and fusion

[0079] In order to fuse feature maps of different scales, upsampling techniques (such as bilinear interpolation) are first used to unify all feature maps to the same size, that is, to keep the same size as the input feature map. Then, pixel-by-pixel weighted summation is performed on the unified feature maps to achieve preliminary feature fusion. In the weighted summation process, different weights can be set as needed to further adjust the contribution of features of each scale in the fusion result.

[0080] S4: High-level feature extraction

[0081] The fused feature maps are further refined through a set of convolutional layers to generate high-level features for final classification and positioning. This set of convolutional layers can include multiple 3x3 convolution kernels, each followed by a ReLU activation function to increase the nonlinear expression ability of the network. Through layer-by-layer convolution and activation, the network is able to learn more abstract and advanced feature representations, which are crucial for the classification and positioning tasks of the target.

[0082] The adaptive feature fusion module of the present invention realizes multi-scale feature extraction through the ASPP module, enhances key features by using the attention mechanism, realizes feature fusion through upsampling and pixel-by-pixel weighted summation, and finally refines high-level features through the convolution layer, which provides a powerful feature representation capability for the intelligent annotation of marine images and improves the recognition and positioning accuracy of targets.

[0083] Embodiment 2:

[0084] When constructing a marine target recognition model, in order to improve the accuracy of the model's bounding box regression of targets in marine images, the present invention uses Smooth L1 loss instead of the original L2 loss as the bounding box regression loss. Smooth L1 loss is more robust to outliers and can effectively reduce the impact of large errors on model training, making the model more stable during training.

[0085] In the specific implementation, for each predicted bounding box, the difference between it and the real bounding box is calculated, including the offset of the center point coordinates and the scaling ratio of the width and height. Then, these difference values ​​are input into the Smooth L1 loss function to calculate the loss value. The definition of the Smooth L1 loss function is as follows:

[0086]

[0087] Where x is the difference between the predicted value and the true value. When |x|<1, the loss is 0.5x 2 ; When |x|≥1, the loss is |x|-0.5. During model training, by minimizing the Smooth L1 loss, the predicted bounding box can be made closer to the true bounding box, thereby improving the accuracy of target positioning.

[0088] In order to further improve the classification accuracy of the model for targets in marine images, the present invention combines the Dice coefficient loss and the cross entropy loss in the classification loss. The Dice coefficient loss can measure the overlap between the predicted segmentation result and the true label, and is particularly sensitive to the segmentation of the target edge, while the cross entropy loss focuses on the accuracy of the overall classification. The combination of the two can complement each other's advantages and improve the classification performance of the model.

[0089] In the specific implementation, for each predicted classification result, the Dice coefficient loss between it and the true label is first calculated. The Dice coefficient loss is defined as:

[0090]

[0091] Among them, A is the predicted segmentation result, B is the true label, |·| represents the number of elements in the set, and by minimizing Dice Loss, a high degree of overlap between the predicted result and the true label is promoted; the smaller the value of the Dice coefficient loss, the higher the overlap between the predicted segmentation result and the true label. At the same time, the cross entropy loss between the predicted classification result and the true label is calculated. The cross entropy loss is defined as:

[0092]

[0093] Where N represents the number of categories, y i represents the indicator value (0 or 1) of the i-th category in the true label, p i Represents the probability of being predicted as the i-th category; the smaller the value of the cross entropy loss, the smaller the overall difference between the predicted classification result and the true label.

[0094] Finally, the Dice coefficient loss and the cross entropy loss are weighted and summed to obtain the final classification loss. The formula for the weighted summation is:

[0095] Total Loss=λ×Dice Loss+(1-λ)×Cross Entropy Loss

[0096] Among them, λ is a hyperparameter used to adjust the relative importance of Dice coefficient loss and cross entropy loss in the total classification loss. During the model training process, by minimizing the total classification loss, the model can be prompted to more accurately identify the target category in the marine image.

[0097] Through the above specific implementation manner, in the intelligent annotation method of marine images of the present invention, by optimizing the loss function, the positioning accuracy and classification accuracy of the model for marine targets are improved.

[0098] Embodiment 3:

[0099] In order to eliminate noise in marine images and improve image quality, the present invention uses Gaussian filtering to smooth the images. Gaussian filtering is a linear smoothing filter, the core idea of ​​which is to use Gaussian function to calculate the weighted average of each pixel in the image and other pixels in its neighborhood, thereby replacing the original pixel value to achieve a denoising effect.

[0100] In the specific implementation, first define the Gaussian function G(x,y), and its calculation formula is:

[0101]

[0102] Among them, (x, y) represents the two-dimensional coordinates in the image, σ is the standard deviation of the Gaussian distribution, which determines the width of the Gaussian function, that is, the smoothness of the filter. e is the base of the natural logarithm, and π is the circumference of a circle.

[0103] Then, for each pixel (i, j) in the image, a neighborhood of size (2k+1)x(2k+1) (k is a positive integer) is selected with it as the center, and the Gaussian function values ​​of all pixels in the neighborhood are calculated and used as weights. Then, using these weights and the grayscale values ​​of the corresponding pixels in the neighborhood, the weighted average is calculated as the new grayscale value of the pixel after denoising.

[0104] By traversing all pixels in the image and applying the above formula, the denoised sea image can be obtained.

[0105] In order to improve the visual effect of marine images, the present invention performs contrast enhancement processing on the images. The specific implementation steps are as follows:

[0106] The brightness histogram of the input marine image is calculated and the number of pixels at each gray level is counted to understand the distribution of image brightness.

[0107] According to the histogram distribution, determine the minimum value I of the image brightness min and the maximum value I max , which is usually achieved by finding the minimum and maximum gray level of the number of non-zero pixels in the histogram.

[0108] Apply the linear stretching algorithm to map the original grayscale value I(x,y) to the new grayscale value I ′ (x, y). Traverse the mapped image and update the grayscale value of each pixel to I ′ (x,y), and obtain the contrast-enhanced sea image.

[0109] In order to simplify the image data and reduce the computational complexity, the present invention performs grayscale processing on the marine image. The specific implementation method is as follows:

[0110] a. Select the weighted method to convert the color image into a grayscale image. The weighted method takes into account the different sensitivities of the human eye to the three primary colors of red, green, and blue, so different weights are used for weighting. The weighted formula used in the present invention is:

[0111] GRAY=0.3×R+0.59×G+0.11×B

[0112] Among them, R, G, and B represent the components of a pixel in a color image in the red, green, and blue color domains respectively.

[0113] b. Read the image data of the marine image, traverse each pixel in the image, and obtain the RGB color value corresponding to each pixel.

[0114] c. Apply the above color value conversion formula to convert the RGB color value of each pixel to obtain the corresponding grayscale value.

[0115] d. Apply the converted grayscale value to each pixel to generate a grayscale image of the sea image. In this way, the original color image is converted into a grayscale image, providing simplified data input for subsequent processing steps.

[0116] Embodiment 4:

[0117] The present invention provides an image recognition system based on big data technology, which integrates multiple functional modules to realize the complete process from image acquisition to target recognition and annotation; the system includes:

[0118] Image acquisition module, the image acquisition module is the input end of the system, equipped with high-resolution camera equipment, such as high-definition cameras or photography devices carried by drones. These devices can capture marine image data containing target objects, such as ships, buoys, marine life, etc. The camera equipment transmits the collected image data to the subsequent processing module through a suitable interface (such as USB, HDMI, network interface, etc.). In order to ensure image quality, the camera equipment may also have functions such as automatic focus and exposure adjustment.

[0119] The preprocessing module receives the image data transmitted by the image acquisition module and performs a series of preprocessing operations on it to improve the image quality and the accuracy of subsequent recognition. These operations include but are not limited to image denoising, contrast enhancement and grayscale processing.

[0120] Model building and training module: The model building and training module is responsible for building and training the maritime target recognition model. The specific implementation is as follows:

[0121] Model construction: Based on the improved DeepLabV3+ network architecture, a maritime target recognition model is constructed. The model has an embedded adaptive feature fusion module that can automatically fuse features of different scales and levels to improve the recognition ability of the model. At the same time, the loss function is optimized, such as cross entropy loss, focal loss, etc., to further improve the recognition accuracy.

[0122] Model training: Prepare a training data set containing offshore target objects, including positive samples (containing targets) and negative samples (excluding targets). Train the model by adjusting model parameters (such as learning rate, batch size, number of iterations, etc.) and formulating training strategies (such as early stopping, learning rate decay, etc.) to enable it to accurately identify offshore target objects.

[0123] The target recognition and annotation module receives the image data processed by the preprocessing module and uses the trained marine target recognition model to perform target recognition. The model will output the annotation results containing the target category and location information, such as the category label of the target object, the bounding box coordinates, etc.

[0124] The control and interface module is responsible for the coordination and control between the modules of the system. It is responsible for the transmission of image data, the configuration of model parameters, the output of recognition results, and the interactive interface with external systems. Through this module, users can easily control the operation of the entire system, obtain recognition results, and pass the results to other systems or applications for further processing.

[0125] The present invention also provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned image recognition method is implemented. Specifically, the electronic device completes tasks such as image acquisition, preprocessing, model building and training, target recognition and labeling, and system control through the collaborative work of its processor and memory.

[0126] In addition, the present invention also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned image recognition method is implemented. The storage medium can be any medium that can store a computer program and enable it to be read and executed by a processor, such as a disk, an optical disk, a USB flash drive, an SD card, etc. By connecting the storage medium to an electronic device (such as a computer, a server, etc.) and running the computer program stored thereon, the intelligent annotation function of the marine image can be realized.

[0127] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0128] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An image recognition method based on big data technology, characterized in that: The method comprises: Step 1: Collecting marine image data: using high-resolution camera equipment to capture marine image data containing the target object; Step 2: Preprocessing the marine image data: preprocessing the captured marine image data, including image denoising, contrast enhancement and grayscale processing; Step 3: Build a maritime target recognition model: Build a maritime target recognition model based on the improved DeepLabV3+ network architecture. The model is used to identify and classify specific targets in maritime images. The model has an embedded adaptive feature fusion module and optimizes the loss function in the original DeepLabV3+ network. Step 4: Identify and label maritime targets: Input the maritime image preprocessed in step 2 into the trained maritime target recognition model, perform the target recognition task, and output the labeling results containing the target category and location information.

2. The image recognition method based on big data technology according to claim 1, characterized in that: The implementation steps of the adaptive feature fusion module include: S1: The input features are processed at multiple scales through the ASPP module to generate feature maps of different scales; S2: Apply the attention mechanism to the feature maps of each scale and assign different weights to features of different regions; S3: After unifying all feature maps to the same size using upsampling technology, perform pixel-by-pixel weighted summation; S4: The fused feature map is further refined through the convolutional layer to generate high-level features for final classification and positioning.

3. The image recognition method based on big data technology according to claim 1, characterized in that: The optimized loss function includes: introducing Smooth L1 loss to replace the original L2 loss as the bounding box regression loss, which is defined as: Where x is the difference between the predicted value and the true value. When |x|<1, the loss is 0.5x 2 ; When |x|≥1, the loss is |x|-0.

5.

4. The image recognition method based on big data technology according to claim 3 is characterized in that: The optimized loss function further includes: using a combination of Dice coefficient loss and cross entropy loss as classification loss, and the Dice coefficient loss is defined as: Among them, A is the predicted segmentation result, B is the true label, |·| represents the number of elements in the set, and by minimizing DiceLoss, a high overlap between the predicted results and the true labels is promoted; at the same time, the cross entropy loss between the predicted classification results and the true labels is calculated. The cross entropy loss is defined as: Among them, N represents the number of categories, y i represents the indicator value of the i-th category in the true label, p i Represents the probability of prediction as the i-th category; The Dice coefficient loss and the cross entropy loss are weighted and summed to balance the two losses. The calculation formula is: Total Loss=λ×Dice Loss+(1-λ)×Cross Entropy Loss Among them, λ is a hyperparameter used to adjust the relative importance of each part of the loss.

5. The image recognition method based on big data technology according to claim 1, characterized in that: The image denoising adopts Gaussian filtering to eliminate the noise in the image, and replaces the value of each pixel in the image with the weighted average of the pixel values ​​in its neighborhood. The calculation formula is: Where G(x,y) is the value of the Gaussian function at the two-dimensional coordinate (x,y) of the image, x and y are the coordinates in two-dimensional space, σ is the standard deviation of the Gaussian distribution, e is the base of the natural logarithm, and π is the circumference of a circle.

6. The image recognition method based on big data technology according to claim 1, characterized in that: The contrast enhancement step in preprocessing the marine image data specifically includes: Calculate the brightness histogram of the input marine image and count the number of pixels at each gray level; According to the histogram distribution, determine the minimum value I of the image brightness min and the maximum value I max ; Apply the linear stretching algorithm to map the original grayscale value I(x,y) to the new grayscale value I′(x,y); The mapped image is traversed, and the grayscale value of each pixel is updated to I′(x, y) to obtain a contrast-enhanced marine image.

7. The image recognition method based on big data technology according to claim 1, characterized in that: The method for grayscale processing of marine image data is: a. Select the weighted method to convert the color image into a grayscale image, and use the formula GRAY = 0.3*R + 0.59*G + 0.11*B to calculate the grayscale value. R, G, and B in the formula represent the components of the pixel in the red, green, and blue color domains respectively; b. Read the image data of the sea image, traverse each pixel in the image, and obtain the RGB color value corresponding to each pixel; c. Apply the color value conversion method, call the formula in step a to convert the RGB color value of each pixel obtained in step b to obtain the corresponding grayscale value; d. Apply the converted grayscale value to each pixel to generate a grayscale image of the sea image.

8. An image recognition system based on big data technology, characterized in that: The system comprises: Image acquisition module: equipped with high-resolution camera equipment to capture marine image data containing target objects and transmit the acquired image data to subsequent processing modules; Preprocessing module: receives the image data transmitted by the image acquisition module and performs image denoising, contrast enhancement and grayscale processing; Model building and training module: Based on the improved DeepLabV3+ network architecture, a maritime target recognition model is built. The model has an embedded adaptive feature fusion module and optimizes the loss function to improve recognition accuracy. This module is also responsible for the model training process, including the preparation of data sets, adjustment of model parameters, and formulation of training strategies. Target recognition and labeling module: receives the image data processed by the preprocessing module, uses the trained marine target recognition model to perform target recognition, and outputs labeling results containing target category and location information; Control and interface module: responsible for the coordination and control between the various modules of the system, including the transmission of image data, the configuration of model parameters, the output of recognition results, and the interactive interface with the external system.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the image recognition method according to any one of claims 1 to 7 is implemented.

10. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the image recognition method according to any one of claims 1 to 7 is implemented.

Images