An image recognition method and system for style transfer of aquatic organisms

Abstract

The application provides a kind of style migration of aquatic organism image recognition method and system, it is related to image recognition technical field.The method includes the following steps: obtaining aquatic organism image in aquaculture environment;Disease characteristic extraction is carried out to the obtained aquatic organism image, and disease characteristic information is obtained;According to disease characteristic information, generate disease characteristic indication map;The style migration operation is executed to the obtained aquatic organism image, and the aquatic organism image after style migration is obtained;The aquatic organism image after style migration is superimposed with disease characteristic indication map and is displayed, and the superimposed display image is obtained;Using superimposed display image, disease identification is carried out to aquatic organism.The method of the application makes up the defect that traditional style migration may cause disease information loss, improves the accuracy of identification and the explainability of result, and then enhances the value of system in early warning and artificial decision.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and more specifically, to an image recognition method and system for style transfer of aquatic organisms. Background Technology

[0002] In modern aquaculture facilities, automated image acquisition and analysis systems are typically deployed to achieve continuous monitoring of the health status of aquatic organisms and early disease warning. These systems acquire image data of aquatic organisms in the aquaculture ponds using underwater optical imaging equipment and transmit it to a central image processing unit for disease identification. The identification model is usually trained on images taken under standard laboratory conditions. These training images are characterized by clean backgrounds, uniform lighting, clear water, relaxed organism postures, and clearly defined disease features, forming a target domain with a high signal-to-noise ratio.

[0003] However, there are significant style differences between real-world aquaculture images and laboratory training samples. Aquaculture water often contains suspended particulate matter, uneaten feed, excrement, and algae, leading to turbidity and consequently reducing the image signal-to-noise ratio and contrast. Lighting conditions vary with time, weather, and equipment location, resulting in uneven lighting, reflections, or shadows. Furthermore, the swimming postures, colony densities, and shooting angles of aquatic organisms also vary. These factors collectively contribute to the lower quality of real-world images compared to laboratory images. Directly inputting real-world images into a recognition model trained on laboratory images will significantly decrease recognition accuracy.

[0004] To bridge this domain gap, existing technologies typically employ image style transfer methods. This method uses a pre-defined style transfer model to perform a single style transfer operation on all acquired field images, such as overall desaturation, brightness equalization, contrast enhancement, or color correction. The aim is to make the style of aquaculture environment images resemble that of laboratory environment images, thereby standardizing image input and improving the efficiency of existing recognition models.

[0005] However, existing image style transfer methods face multiple challenges in practical applications. First, the visual characteristics of diseases in aquatic organisms, especially in the early stages, are often subtle and atypical. For example, early parasitic infections may lead to increased mucus secretion on the body surface, while early bacterial infections may manifest as slight adhesion of gill filaments or localized congestion. In the complex and variable aquaculture environment, these atypical features can exhibit complex and diverse visual patterns due to the influence of water media, lighting conditions, and the organism's own physiological activities. The main goal of existing style transfer methods is to "standardize" field images to a laboratory style. This may cause these unique, diagnostically valuable visual information in complex environments to be smoothed, blurred, or eliminated during the conversion process, thereby reducing the detection capability of subsequent disease identification models for early, subtle, or special diseases and increasing the false negative rate.

[0006] Secondly, the visual features of diseases evolve dynamically as the disease progresses. Style transfer methods with fixed parameters struggle to adapt to these dynamic changes. Parameters optimized for weak early features may lead to overprocessing when dealing with large-area lesions in the later stages; while parameters optimized for significant late-stage lesions may fail to effectively recover subtle color changes or texture anomalies from the early stages, resulting in insufficient early warning capabilities. This lack of adaptability to the disease's evolutionary stages limits the system's recognition performance throughout the entire disease cycle.

[0007] Furthermore, existing style transfer methods typically optimize image visual quality or style similarity, rather than directly targeting performance in subsequent disease identification tasks. Even if the converted image appears highly similar to laboratory images to the human eye, if this similarity is achieved by compromising the integrity of disease features or introducing artifacts detrimental to the identification model, then such conversion is inefficient or has a negative impact on disease identification tasks. For example, over-sharpening may introduce edge artifacts, and over-smoothing may erase minute lesions. This "style" matching is not equivalent to "feature" matching, and may even lead to "feature mismatch," preventing the identification model from fully utilizing the converted image information and thus reducing identification accuracy.

[0008] Ultimately, automated disease identification systems typically require manual intervention or verification. If the style transfer process excessively modifies the original image or introduces incomprehensible artifacts, weakening the visual correlation between the converted image and the original lesions, or obscuring some key information in the original image, then farmers will be confused during manual review and will find it difficult to determine the reliability of the identification results. This not only increases the time and cost of manual verification, but more importantly, it reduces farmers' trust in the automated system, weakening the practical value of automated early warning. Summary of the Invention

[0009] The purpose of this invention is to provide an image recognition method and system for style transfer in aquatic organisms, which overcomes the shortcomings of traditional style transfer that may lead to the loss of disease information, improves the accuracy of recognition and the interpretability of results, and thus enhances the value of the system in early warning and human decision-making.

[0010] In a first aspect, the present invention provides an image recognition method for style transfer in aquatic organisms, comprising the following steps:

[0011] Acquire images of aquatic organisms in aquaculture environments;

[0012] Disease feature extraction was performed on the acquired aquatic organism images to obtain disease feature information;

[0013] Based on the disease characteristic information, generate a disease characteristic indicator map;

[0014] A style transfer operation is performed on the acquired aquatic organism images to obtain style-transferred aquatic organism images. The style transfer operation controls the image processing of the diseased areas based on the disease feature information to preserve the integrity of the disease features.

[0015] The style-transferred aquatic organism images are overlaid with disease feature indicator maps to obtain an overlaid display image;

[0016] Disease identification of aquatic organisms is achieved by using overlaid images.

[0017] The image recognition method for style transfer of aquatic organisms provided by this invention effectively avoids the loss or distortion of key, minor, and atypical disease features during the conversion process. At the same time, it prevents the introduction of artifacts unfamiliar to the recognition model, which would lead to a decrease in recognition accuracy. Furthermore, it ensures that the converted image can provide clear and intuitive visual evidence of lesions to support efficient manual review and enhance system reliability.

[0018] Secondly, the present invention provides an image recognition system for style transfer in aquatic organisms, comprising:

[0019] The acquisition module is used to acquire images of aquatic organisms in the aquaculture environment;

[0020] The extraction module is used to extract disease features from the acquired aquatic organism images to obtain disease feature information;

[0021] The generation module is used to generate disease feature indication maps based on disease feature information;

[0022] The style transfer module is used to perform style transfer operations on the acquired aquatic organism images to obtain style-transferred aquatic organism images. The style transfer operation controls the image processing of the diseased area based on the disease feature information to preserve the integrity of the disease features.

[0023] The overlay module is used to overlay the style-transferred aquatic organism images with the disease feature indicator maps to obtain an overlay display image;

[0024] The image recognition module is used to identify diseases in aquatic organisms by overlaying images.

[0025] As can be seen from the above, the image recognition method for style transfer of aquatic organisms provided by the present invention effectively solves the problems in the prior art that may lead to loss of disease information, decreased recognition accuracy, and difficulty in manual verification due to image style transfer, by using a strategy of parallel image processing and result fusion.

[0026] By extracting minute, atypical disease features from the original images before or in parallel with style transfer and generating indicator maps, it is ensured that the visual information of diseases with diagnostic value that is unique to complex aquaculture environments is not smoothed, blurred, or eliminated by the style transfer process, which significantly improves the sensitivity of early disease warning.

[0027] The style-transferred images retain key disease features while providing standardized inputs that are closer to the training samples for the automated identification model. At the same time, by overlaying the original disease feature indicator map, visual cues are provided to assist the identification system in making judgments, making up for possible feature mismatch problems, thereby improving the overall accuracy of automated disease identification.

[0028] The overlay display module intuitively presents standardized images of organisms and precise markings of disease clues in the original images. When farmers conduct manual reviews, they can clearly and quickly locate and understand the lesion areas, greatly reducing the complexity and time cost of the review process. This significantly enhances farmers' trust in the automated disease early warning system and their willingness to adopt it.

[0029] The disease feature extraction module of this solution can capture weak atypical features in the early stage of disease, and the style transfer process avoids destructive processing of these features, enabling the system to better adapt to the dynamic evolution of disease from early to late stage and achieve effective monitoring of the entire disease course.

[0030] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings. Attached Figure Description

[0031] Figure 1 This is a flowchart of an image recognition method for style transfer of aquatic organisms provided in an embodiment of the present invention.

[0032] Figure 2 This is a schematic diagram of an image recognition system for style transfer of aquatic organisms provided in an embodiment of the present invention.

[0033] Label Explanation:

[0034] 100. Acquisition Module; 200. Extraction Module; 300. Generation Module; 400. Style Transfer Module; 500. Overlay Module; 600. Image Recognition Module. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0036] Reference Appendix Figure 1 This invention provides an image recognition method for style transfer in aquatic organisms, comprising the following steps:

[0037] Acquire images of aquatic organisms in aquaculture environments;

[0038] Disease feature extraction is performed on the acquired aquatic organism images. Disease feature extraction includes identifying local brightness, color, texture or morphological changes related to diseases in the aquatic organism images to obtain disease feature information.

[0039] Based on the disease characteristic information, generate a disease characteristic indicator map;

[0040] A style transfer operation is performed on the acquired aquatic organism images to obtain style-transferred aquatic organism images. The style transfer operation controls the image processing of the diseased areas based on the disease feature information to preserve the integrity of the disease features.

[0041] The style-transferred aquatic organism images are overlaid with disease feature indicator maps to obtain an overlaid display image;

[0042] Disease identification of aquatic organisms is achieved by using overlaid images.

[0043] Disease feature extraction refers to the identification of disease-related visual anomalies from aquatic organism images. This can be achieved using image processing algorithms, machine learning models, or deep learning networks, such as threshold-based segmentation, texture analysis, color space conversion, or convolutional neural network feature extraction. Its primary purpose is to identify potential disease regions and their visual manifestations within the image. Disease feature information refers to the output of the disease feature extraction process, containing the location, extent, type, severity, or visual manifestation parameters of disease regions in the image. This information can be represented using pixel coordinates, region masks, classification labels, or quantitative indicators, such as the bounding box of the lesion region, the pixel set of the lesion region, disease type labels, or lesion severity scores. Its main purpose is to provide a data foundation for subsequent style transfer control and disease indicator map generation. Disease feature indication map refers to an image or data structure generated based on disease feature information to intuitively represent the disease area and its attributes. It can be implemented using binary mask map, heat map, color-coded map or overlay layer. For example, the disease area is highlighted with a specific color, different disease types are marked with different textures, or the severity of the disease is indicated by changes in transparency. Its main purpose is to provide a reference for the location and attributes of the disease, which is convenient for subsequent overlay display and manual verification. Style transfer operations control the image processing of diseased areas based on disease feature information to preserve the integrity of disease features. This refers to applying differentiated processing strategies to areas containing diseased features during image style transfer, using pre-extracted disease feature information. This prevents disease features from being smoothed, blurred, or eliminated during style transfer. This can be achieved through local weight adjustment, region protection masks, weighted feature loss functions, or adaptive transfer parameter adjustments. For example, reducing the style transfer intensity in diseased areas, applying higher reconstruction loss to disease feature pixels, or adjusting local brightness and contrast parameters based on disease features. The main goal is to ensure that style transfer improves the overall visual effect of the image while maintaining the original state and recognizability of disease diagnostic information. Overlay display refers to the fusion of style-transferred aquatic organism images with disease feature indicator maps. This can be achieved through image blending, layer overlay, or transparency fusion. For example, the disease indicator map can be overlaid on the style-transferred image in a semi-transparent manner, the boundaries or colors of the disease area can be drawn directly on the image, or pixel value weighted fusion can be used. The main purpose is to provide an image that simultaneously includes visual optimization effects and clear disease annotations, facilitating automated identification and manual verification.

[0044] The working principle of this invention is that the system acquires raw images of aquatic organisms from underwater equipment. Subsequently, the image data is fed into two parallel processing paths:

[0045] The first path (disease feature extraction and indication): This module performs specialized disease feature extraction on the original image. It focuses on identifying subtle, atypical visual cues of disease that may be overlooked or disrupted by existing style transfer methods in complex aquaculture environments, such as localized abnormal gloss, fine roughness, gill filament adhesion, or localized congestion. This process suppresses the effects of environmental disturbances such as water turbidity and uneven lighting. The extracted disease features are used to generate a disease feature indication map that accurately marks the location and extent of the disease.

[0046] The second path (controlled style transfer) involves performing a style transfer operation on the original image to convert its overall visual style into a standardized style similar to the lab training images, thus facilitating subsequent processing by automated recognition models. Crucially, this style transfer process avoids destructive processing such as excessive smoothing or sharpening of the disease feature regions identified in the first path, preserving the integrity of the original disease information to the greatest extent possible.

[0047] Finally, the overlay display module uses the style-transferred image output from the second path as the bottom layer, and overlays the disease feature indicator map generated by the first path as a transparent layer on top of it. This overlay display method allows automated identification systems or human reviewers to simultaneously obtain intuitive markings of disease clues in both the standardized style image of the organism and the original image. In this way, the present invention overcomes the deficiency of traditional style transfer that may lead to the loss of disease information, improves the accuracy of identification and the interpretability of results, and thus enhances the value of the system in early warning and human decision-making.

[0048] The core innovation of this application lies in introducing a disease feature recognition and protection mechanism during the image style transfer process. Specifically, the image processing of the diseased area is controlled based on the disease feature information to preserve the integrity of the disease features. Furthermore, the style-transferred image is overlaid with the disease feature indicator image, thereby solving the problem that existing style transfer methods may lead to the loss or distortion of disease features. This achieves the effects of improving disease recognition accuracy, enhancing early warning capabilities, and facilitating manual review.

[0049] Specifically, this method aims to address the problem that existing image style transfer may lead to the loss or distortion of disease features in aquaculture environments. By introducing a disease feature protection mechanism during the style transfer process and combining it with disease indicator maps for overlay display, the accuracy and reliability of disease identification can be improved, while also facilitating manual verification.

[0050] First, the system acquires images of aquatic organisms in the aquaculture environment. These images are affected by environmental factors such as water turbidity, uneven lighting, and suspended matter.

[0051] Subsequently, before style transfer, the system extracts disease features from the original image, identifying local brightness, color, texture, or morphological changes related to the disease, thus obtaining disease feature information. This step identifies visual cues related to the disease in the image, providing protected targets for subsequent style transfer operations.

[0052] Next, based on the extracted disease feature information, the system generates a disease feature indicator map. This indicator map is a visual representation of the disease features, providing a disease reference for subsequent overlay display.

[0053] Then, the system performs style transfer on the acquired aquatic organism images. Unlike existing technologies that perform a single style transfer on all acquired images, this method controls the image processing of diseased areas based on disease feature information during style transfer to preserve the integrity of disease features. By utilizing previously extracted disease feature information, the system can identify and protect diseased areas in the image, avoiding excessive smoothing, blurring, or elimination of these visual cues during operations such as denoising, deturbination, and brightness equalization. This targeted control ensures that style transfer improves the overall visual quality of the image while preserving the original appearance of the disease.

[0054] Subsequently, the system overlays the style-transferred aquatic organism images with the disease feature indicator maps to obtain an overlaid image. This step combines the style-optimized image, which retains the disease features, with the indicator map that clearly marks the location and attributes of the disease, resulting in a visually appealing image that also indicates the diseased area.

[0055] Ultimately, the system utilizes overlaid images to identify diseases in aquatic organisms. Because the overlaid images combine the quality advantages of style-transferred images with clear annotations of disease feature indicators, they provide comprehensive and accurate input to the disease identification model, thereby improving the accuracy, robustness, and early warning capabilities of the identification process.

[0056] As one embodiment, the solution of this application is implemented as follows:

[0057] First, images of fish in aquaculture ponds are acquired using underwater cameras, and the image data is transmitted to an image processing server via the network.

[0058] In the image processing server, a deep learning-based image segmentation network, such as U-Net or Mask R-CNN, is used to extract disease features from the acquired fish images. After training, this network can identify visual disease features on the fish's surface, such as local color anomalies, rough texture, and morphological distortions, and output pixel-level masks of diseased areas and disease type classification results as disease feature information.

[0059] Subsequently, based on these disease characteristic information, a disease feature indication map is generated. For example, the pixel mask of the diseased area is converted into a binary map with the same size as the original image, where the pixel value of the diseased area is 1 and the pixel value of the non-diseased area is 0.

[0060] Next, style transfer is performed on the original fish images. This style transfer model can employ a Generative Adversarial Network (GAN) architecture, such as CycleGAN or StarGAN. During style transfer, the generator loss function of the GAN is modified using previously generated disease feature information (pixel masks). For example, higher perceptual loss weights or content loss weights are applied to pixels in diseased areas to ensure that the original pixel values ​​and structural information of diseased areas are preferentially preserved during style transfer. Simultaneously, non-diseased areas are processed according to style transfer strategies to remove environmental interference such as water turbidity and uneven lighting.

[0061] Next, the style-transferred fish image is overlaid with the disease feature indicator map. For example, the disease feature indicator map (binary mask) is used as a transparency channel and overlaid on the style-transferred image as a red semi-transparent overlay, so that the diseased area is highlighted in red, while not completely obscuring the details of the underlying image.

[0062] Finally, the overlaid image is input into a pre-trained disease identification classifier, such as ResNet or EfficientNet, which uses the visual features of the image and the overlaid disease indication information to identify the type and severity of disease in the fish.

[0063] By employing the aforementioned solution, this application addresses the problem of lost or distorted aquatic organism disease characteristic information caused by the pursuit of image style standardization in existing image style transfer methods within aquaculture environments. This method ensures the integrity of disease characteristics by specifically protecting diseased areas during the style transfer process, thereby improving the accuracy of subsequent disease identification. Furthermore, the overlay display of disease characteristic indicator maps enhances the intuitiveness and interpretability of disease identification results, facilitates manual verification, and ultimately increases the value of the automated disease early warning system and the trust of aquaculture personnel in the system.

[0064] In some embodiments described above, a method for extracting disease features from acquired aquatic organism images is proposed to obtain disease feature information. This feature extraction can specifically involve applying general image processing algorithms, such as edge detection, color histogram analysis, or simple threshold segmentation, to identify potentially abnormal regions in the image. This can initially screen out potential disease areas. However, in actual aquaculture environments, the aquatic environment is complex and variable. Disease features in aquatic organisms, especially in the early stages of disease, often manifest as subtle and atypical visual changes, such as minor anomalies in local brightness, color, texture, or morphology. These subtle features are easily interfered with or masked by environmental factors such as water turbidity, uneven lighting, and suspended matter, making it difficult for general feature extraction methods to accurately and comprehensively capture this crucial disease information. If the disease feature extraction is not refined enough or lacks specificity, the subsequently generated disease feature indication map may be inaccurate. Furthermore, during style transfer, these crucial and subtle disease features may be smoothed, blurred, or misjudged as background noise and lost, thus affecting the accuracy of final disease identification and early warning capabilities.

[0065] In some embodiments, the step of performing a style transfer operation on the acquired aquatic organism images to obtain style-transferred aquatic organism images includes:

[0066] Based on the disease characteristics information, determine the disease areas in the aquatic organism images that require disease feature protection and the protection intensity parameters;

[0067] Local image feature analysis was performed on non-disease areas adjacent to diseased areas in aquatic organism images to obtain local image feature information of non-disease areas;

[0068] Based on the local image features of the diseased area, protection intensity parameters, and non-diseased areas, the local transformation parameters of the style transfer operation are adaptively adjusted.

[0069] By using adaptively adjusted local transformation parameters, a style transfer operation is performed to obtain style-transferred images of aquatic organisms.

[0070] This method, when performing style transfer operations on acquired aquatic organism images, aims to address the problem of easily disturbed and difficult-to-preserve disease features in complex aquaculture environments by introducing a refined local control mechanism. The diseased region refers to a specific set of pixels or image area in the aquatic organism image identified as exhibiting signs of disease. Its determination can be based on disease feature information obtained in previous steps, such as identifying areas with abnormal brightness, color, texture, or morphological changes through image segmentation algorithms. The protection strength parameter is a quantitative indicator used to indicate the degree to which the original features of the diseased region are preserved during style transfer. This parameter can be a continuous value, such as from 0 to 1, where higher values ​​indicate stronger protection, or a discrete level, such as "low," "medium," or "high" protection. Its setting can be based on the type, severity, clarity of lesions, or their importance in the image. Local image feature analysis refers to the process of quantifying and describing the visual attributes of specific small areas in the image. For non-disease areas adjacent to diseased areas, this analysis can extract their color distribution, texture patterns, brightness gradients, edge information, or statistical characteristics to characterize the visual properties of the local water environment, lighting conditions, or background. This analysis can employ traditional image processing algorithms, such as gray-level co-occurrence matrices and local binary patterns, or deep learning-based feature extraction methods. Local transformation parameters refer to specific variables or weights that control the transformation of image pixels or features in style transfer algorithms. In global style transfer, these parameters are usually fixed, but in this scheme, they are designed to be differentially adjusted for different local regions of the image. These parameters may include, but are not limited to, the weights of content loss and style loss, feature map scaling factors, color mapping rules, or the intensity of texture synthesis. Adaptive adjustment means that the modification of local transformation parameters is dynamic, non-linear, and automatic based on the specific local characteristics of the input image (i.e., diseased areas, protection strength parameters, and local image feature information of non-disease areas). This adjustment enables the style transfer process to intelligently adapt to the needs of different regions in the image. For example, it can maintain the original features in diseased areas while actively performing style transfer in non-diseased areas, in order to achieve a balance between the unity of the overall style and the integrity of local features.

[0071] Specifically, before performing style transfer, the system accurately identifies diseased areas in the aquatic organism image that require disease feature protection based on the disease feature information obtained in the previous steps. Simultaneously, the system assigns a corresponding protection strength parameter to these diseased areas. This parameter allows the system to flexibly set the preservation level of disease features based on the actual situation of the disease, such as its type, severity, or visual clarity, thereby avoiding unnecessary modifications to key diagnostic information. Next, to ensure a natural transition between diseased and non-diseased areas during style transfer and to fully consider the influence of the local environment on the visual representation of disease features, the system performs local image feature analysis on non-diseased areas adjacent to the diseased areas in the aquatic organism image. This analysis obtains local image feature information of the non-diseased areas, reflecting the visual characteristics of background factors such as the aquatic environment and lighting conditions. This background information is crucial for understanding how disease features are presented in a specific environment and provides the necessary context for subsequent adjustments to local transformation parameters. Building upon this foundation, the core of this method lies in adaptively adjusting the local transformation parameters of the style transfer operation based on the determined diseased areas, protection intensity parameters, and local image feature information obtained from non-diseased areas. This adjustment mechanism comprehensively considers the characteristics of the disease itself, the desired level of protection, and the visual characteristics of the surrounding environment. For example, for diseased areas, the content preservation weight in the style transfer algorithm can be dynamically adjusted according to the protection intensity parameter, making it more inclined to maintain the integrity of the original pixels or features; while for non-diseased areas, the intensity or method of style transfer can be adjusted according to their local image feature information, making it better match the target style while ensuring a smooth transition with the boundary of the diseased area. This adaptive, localized parameter adjustment makes style transfer no longer a globally uniform rigid transformation, but a customized processing for different regions of the image, thus effectively avoiding the loss or distortion of disease features caused by excessive smoothing or sharpening. Finally, using the adaptively adjusted local transformation parameters, the system performs the style transfer operation to generate style-transferred aquatic organism images. Because the style transfer process fully considers the need to protect disease features and the influence of the local environment, the resulting image not only maintains overall stylistic consistency with the target domain image, but more importantly, it effectively and completely preserves internal disease features, even those that are small, early-stage, or have blurred boundaries. This refined style transfer processing provides high-quality, high-fidelity input images for subsequent disease identification steps. By overlaying this optimized style-transferred image with a disease feature indicator map and using it for disease identification, the accuracy and reliability of disease identification can be significantly improved, especially in detecting early-stage diseases in complex aquaculture environments, thereby enhancing the practical value and early warning capabilities of the entire image recognition method.

[0072] In some embodiments, the step of adaptively adjusting the local transformation parameters of the style transfer operation based on the local image feature information of the diseased area, the protection strength parameter, and the non-diseased area includes:

[0073] Image feature analysis is performed on the boundary between diseased and non-diseased areas to identify the feature gradient or feature continuity at the boundary, and confidence information of the diseased area boundary is obtained by combining the disease feature information.

[0074] Based on the local image feature information of non-disease areas, the influence of water environmental factors on image features is quantified to obtain environmental interference feature quantity;

[0075] Based on the diseased area, protection strength parameters, confidence information of the diseased area boundary, and environmental interference features, local transformation rules are generated for different image regions.

[0076] The local transformation parameters of the style transfer operation are adjusted using local transformation rules.

[0077] Image feature analysis refers to obtaining descriptive information about image content by calculating or extracting the attributes of image pixels, such as brightness, color, texture, or shape. This can be achieved using edge detection algorithms, texture analysis algorithms, or deep learning-based feature extraction networks. Feature gradient refers to the rate of change of image pixel values ​​in space, reflecting the intensity and direction of image edges or textures. It can be calculated using the Sobel operator, Prewitt operator, or Canny edge detection algorithm. Feature continuity refers to the smoothness or consistency of pixel attributes within an image region, reflecting the uniformity of the region. It can be evaluated using region growing algorithms or clustering-based image segmentation methods. Confidence information of disease area boundaries refers to a quantitative assessment of the accuracy or reliability of the identified disease area boundaries, which can be represented by probability values, ambiguity indices, or classification scores. Environmental interference features refer to numerical or vector values ​​that quantify the influence of non-disease factors in the aquatic environment on the visual performance of an image. They can be characterized by the overall average brightness, contrast, chromaticity deviation, or intensity of specific frequency components of the image. Among them, local transformation rules refer to specific strategies or mapping relationships that guide the adjustment of style transfer operation parameters for different regions in an image. They can be implemented using lookup tables, conditional functions, or local parameter generation modules based on neural networks.

[0078] This scheme, while performing style transfer operations on aquatic organism images to preserve the integrity of disease features, further refines the adaptive adjustment process of local transformation parameters. Specifically, to overcome the problems of blurred disease area boundaries and aquatic environmental interference, this scheme first performs image feature analysis on the boundary between diseased and non-disease areas. By identifying the feature gradient or feature continuity at the boundary and combining it with existing disease feature information, the system can obtain the confidence information of the disease area boundary. This information reflects the certainty of the disease boundary, providing a reliable basis for subsequent parameter adjustment and avoiding the misprotection of non-disease areas or the over-processing of disease edges. Simultaneously, based on the local image feature information of non-disease areas, this scheme quantifies the influence of aquatic environmental factors on image features, thereby obtaining environmental interference feature quantities. This step aims to isolate visual changes caused by environmental factors such as water turbidity and uneven lighting, distinguishing them from true disease features. By quantifying environmental interference, the system can avoid mistaking environmental noise for disease features and protecting it, or submerging true disease features in environmental noise. Based on this, the system comprehensively utilizes the diseased area, preset protection strength parameters, confidence information of newly acquired diseased area boundaries, and environmental interference features to generate local transformation rules for different image regions. This rule generation is the result of multi-dimensional information fusion. For example, for diseased core areas with clear boundaries and minimal environmental interference, stronger protection strategies can be applied; while for areas with blurred boundaries or significant environmental interference, more intelligent or conservative transformation rules can be adopted to minimize the negative impact of environmental factors while protecting disease features. Finally, these generated local transformation rules are used to adjust the local transformation parameters of the style transfer operation. This means that the style transfer parameters are no longer globally uniform but dynamically adjusted according to the actual situation of each local region in the image. This refined adaptive adjustment enables the style transfer process to more accurately protect the subtle, atypical, or environmentally affected features of the diseased area, while ensuring effective style transfer in other regions. In this way, this scheme, combined with the previous scheme that adaptively adjusts based on the local image features of the diseased area, protection intensity parameters, and non-diseased areas, forms a more robust and accurate style transfer mechanism. This significantly improves the authenticity and identifiability of disease features after style transfer in complex aquaculture environments, providing high-quality input for subsequent disease identification and effectively solving the problem of disease feature loss or decreased identification accuracy caused by blurred boundaries and environmental interference.

[0079] In some embodiments, the step of adaptively adjusting the local transformation parameters of the style transfer operation based on the local image feature information of the diseased area, the protection strength parameter, and the non-diseased area includes:

[0080] Based on the local image feature information of non-disease areas, the correlation pattern between water environmental factors and image features of disease areas is analyzed to obtain environmental correlation features;

[0081] By combining the disease area, protection intensity parameters, and environmental correlation characteristics, the visual patterns of disease characteristics presented in the aquatic environment are identified, and environmental disease pattern information is obtained.

[0082] Based on environmental disease pattern information, an environmental adaptation protection strategy is generated for local transformation parameters of style migration operations;

[0083] By combining the local image features of the diseased area, the protection intensity parameters, and the non-diseased area, and by using an environmental adaptation protection strategy, the local transformation parameters of the style transfer operation are adjusted.

[0084] Environmental correlation features refer to quantitative or descriptive information obtained by analyzing the interaction between aquatic environmental factors and image features of diseased areas. This information can be revealed through statistical analysis, machine learning models, or rule-based reasoning to demonstrate the impact of the environment on the visual representation of diseases. Environmental disease pattern information refers to the unique visual manifestations or patterns of disease features under specific aquatic environments. This information can be described using feature vectors, pattern templates, or parameter sets to depict the visual changes of diseases under different environments. Environmental adaptation protection strategies refer to the adjustment rules or guidelines established for the local transformation parameters of style transfer operations based on the identified environmental disease pattern information. These strategies can use lookup tables, conditional logic, or neural network models to dynamically adjust style transfer parameters to ensure the integrity of disease features.

[0085] This solution optimizes the adaptive adjustment process of local transformation parameters in style transfer operations by introducing a deep understanding of the complex relationship between aquatic environmental factors and disease characteristics, thereby ensuring more effective preservation of the integrity of disease characteristics in complex aquaculture environments. Specifically, firstly, the system analyzes the correlation patterns between aquatic environmental factors and image features of diseased areas based on local image feature information from non-disease-affected areas, thus obtaining environmental correlation features. This process, through analysis of image features from non-disease-affected areas, extracts the visual representation of the current aquatic environment, such as turbidity, color, or lighting conditions. Subsequently, these environmental features are analyzed in depth with the image features of diseased areas to reveal how environmental factors specifically affect the visual presentation of diseases, such as how the color, texture, or brightness of diseases may change under specific lighting or turbidity conditions. By obtaining environmental correlation features, the system can gain a quantitative understanding of the complex relationship between the environment and the visual representation of diseases, providing more insightful environmental perception information for subsequent style transfer and avoiding the blind spots or inadequacies that may result from protection based solely on the characteristics of the diseased area itself.

[0086] Secondly, after obtaining environmental correlation features, the system combines the disease area, protection intensity parameters, and environmental correlation features to identify the visual patterns of disease features in the aquatic environment, thereby obtaining environmental disease pattern information. This step utilizes environmental correlation features, combined with the identified disease area and preset protection intensity parameters, to identify the unique visual patterns of disease features in the current aquatic environment. This means that the system no longer treats disease features as a fixed visual representation, but understands their dynamic changes and unique manifestations in different environments. For example, in turbid water, the faint luster anomaly of early-stage disease may manifest as a specific light and shadow pattern, while in clear water it may manifest as another. By identifying this environmental disease pattern information, the system can more accurately grasp the true visual signals of disease in complex environments, providing more targeted protection targets for subsequent style transfer, ensuring that style transfer does not erase these environmentally specific but diagnostically valuable visual cues of disease in the pursuit of "standardization."

[0087] Furthermore, based on the identified environmental disease pattern information, the system generates an environmental adaptation protection strategy for the local transformation parameters of the style transfer operation. This strategy is specifically designed for the local transformation parameters of the style transfer operation, guiding how the style transfer process adjusts its processing method according to the visual pattern of the disease in the current environment. For example, if the environmental disease pattern information shows that a certain texture feature of the disease is easily blurred in the current environment, then the strategy will instruct the style transfer process to particularly strengthen the preservation of texture details when processing that area. The generation of this strategy makes style transfer no longer a simple denoising or standardization, but an intelligent process that formulates transformation rules that are most conducive to preserving the integrity of disease features based on the influence of the environment on the disease features, thereby avoiding over-processing or under-processing of disease features.

[0088] Finally, the system combines the previously generated environmental adaptation protection strategy with the original disease area, protection intensity parameters, and local image feature information of non-disease areas to make final adjustments to the local transformation parameters of the style transfer operation. This means that the parameter adjustment of style transfer is no longer simply based on region and intensity, but incorporates a deep understanding of the complex relationship between the environment and disease features. By utilizing the environmental adaptation protection strategy, the system can ensure that when performing style transfer, it considers not only the importance of the disease itself, but also how the environment affects the visual presentation of the disease. Thus, during the transformation process, it can more accurately preserve disease features that may be ignored or distorted in complex environments, ultimately improving the integrity and recognizability of disease features in the style-transferred image, providing high-quality input for subsequent disease identification. This scheme, based on the basic scheme (i.e., adaptively adjusting the local transformation parameters of the style transfer operation according to the local image feature information of the disease area, protection intensity parameters, and non-disease areas), further introduces the analysis of the correlation pattern between aquatic environmental factors and disease features, and generates an environmental adaptation protection strategy accordingly. This explicit modeling and utilization of environmental factors allows style transfer parameter adjustments to move beyond blindly pursuing universal style matching. Instead, it enables refined protection of disease visual patterns specific to particular aquatic environments. By identifying and adapting to environment-specific disease visual patterns, this approach avoids unintentionally smoothing, blurring, or eliminating diagnostically valuable disease visual information during style transfer. This ensures that style-transferred images more accurately preserve the integrity of disease features, providing more reliable input for subsequent disease identification and significantly improving the accuracy and early warning capabilities of disease identification in complex aquaculture environments.

[0089] In some embodiments, the step of analyzing the correlation pattern between aquatic environmental factors and image features of diseased areas based on local image feature information of non-disease-affected areas to obtain environmental correlation features includes:

[0090] Environmental factor characterization analysis was performed on local image feature information of non-disease areas to obtain visual characterization features of the water environment;

[0091] Visual features of the disease are extracted from the image feature information of the diseased area to obtain the visual characteristics of the disease.

[0092] Based on the visual characteristics of the aquatic environment and the visual manifestations of diseases, environmental correlation characteristics are obtained.

[0093] Environmental factor characterization analysis refers to processing the visual information contained in non-disease areas of an image to quantify and identify various visual attributes of the aquatic environment. This can be achieved using various techniques, such as analyzing the image's color histogram, brightness distribution, texture complexity, spectral characteristics, or pixel intensity in specific bands to assess environmental parameters like turbidity, chromaticity, illumination uniformity, and suspended solids content. The visual characterization features of the aquatic environment refer to the data set obtained through environmental factor characterization analysis that objectively reflects the visual state of the aquatic environment. This can be represented by a set of quantified parameters, such as water transparency index, color temperature value, mean illumination intensity, mean or variance of specific color channels, or more complex feature vectors, used to describe the visual attributes of the aquatic environment. Disease visual feature extraction refers to the process of identifying and quantifying visual attributes related to the disease itself from image information of the diseased area. This can employ image processing algorithms, such as edge detection, texture analysis, color segmentation, morphological operations, or deep learning models, to identify the shape, size, color, texture, boundary clarity, and internal structure of lesions. The visual characteristics of a disease refer to the data set obtained through the extraction of visual features of the disease, which objectively reflects the visual presentation of the disease in images. These can be represented by quantitative indicators such as the average color value of lesions, texture descriptors, shape parameters, brightness contrast, etc., or by high-dimensional feature vectors extracted by neural networks. Environmental correlation features refer to the data set obtained by comprehensively analyzing the visual representation characteristics of the aquatic environment and the visual characteristics of the disease, which reveals how environmental factors affect the visual presentation pattern of the disease. These can be represented by a set of correction coefficients, mapping relationships, conditional probability distributions, or predictive models to describe the possible shifts, enhancements, or reductions in the visual characteristics of the disease under specific aquatic environments.

[0094] This scheme refines the process of analyzing the correlation patterns between aquatic environmental factors and image features of diseased areas, thereby obtaining more accurate environmental correlation characteristics. First, the system performs environmental factor characterization analysis on local image feature information of non-disease-affected areas. This step aims to independently identify and quantify the visual attributes of the aquatic environment itself from the image, such as turbidity, color, and lighting conditions. By focusing on non-disease-affected areas, interference from disease features in the environmental assessment can be avoided, ensuring that the obtained visual representation characteristics of the aquatic environment are pure and objective. Next, the system extracts disease visual features from the image feature information of diseased areas. This process focuses on the visual manifestation of the disease itself in the image, such as the color, texture, and shape of lesions. By isolating disease features from their complex environment for independent analysis, the obtained visual manifestation characteristics of the disease are inherent to the disease and are original information uncontaminated by the environment. Finally, after obtaining the independent visual representation characteristics of the aquatic environment and the visual manifestation characteristics of the disease, the system performs a comprehensive analysis based on these two types of features to obtain the environmental correlation characteristics. This step-by-step and independent analysis method enables the system to more accurately reveal how aquatic environmental factors specifically affect the visual presentation of diseases. For example, the color of lesions may shift under specific water turbidity, or the texture of lesions may be blurred under specific lighting conditions. This precise environmental correlation provides a solid foundation for subsequent identification of the visual patterns of disease features in aquatic environments, allowing the system to more intelligently generate environmental adaptation protection strategies for local transformation parameters of style transfer operations. In this way, this solution effectively addresses the problem that disease features are easily masked or distorted by environmental factors in complex aquaculture environments, ensuring that style transfer operations have higher accuracy and adaptability in protecting the integrity of disease features. This refined analysis method allows for more accurate consideration of the specific impact of the environment on the visual appearance of diseases when adaptively adjusting the local transformation parameters of style transfer operations, thereby improving the robustness and accuracy of the overall image recognition method.

[0095] In some embodiments, the step of obtaining environmental correlation characteristics based on the visual representation characteristics of the aquatic environment and the visual manifestation characteristics of the disease includes:

[0096] The visual representation features of the aquatic environment are matched with the visual manifestation features of diseases to obtain the matched feature pairs.

[0097] Calculate the difference or correlation index between the matched feature pairs to obtain the feature association measure;

[0098] Based on feature correlation measurement, the degree and direction of influence of aquatic environmental factors on the visual manifestation characteristics of diseases are determined, and environmental impact parameters are obtained and used as environmental correlation features.

[0099] Feature dimension matching refers to aligning image features from different sources or semantic levels, allowing them to be compared or fused on comparable dimensions. This can be achieved using techniques such as feature vector-based projection transformation, semantic label-based mapping, or statistical distribution-based normalization. Dissimilarity or correlation indices are numerical measures used to quantify the similarity or dissimilarity between two or more data points or feature sets. They can be calculated using methods such as Euclidean distance, cosine similarity, Pearson correlation coefficient, mutual information, or dynamic time warping. Feature association metrics are quantitative values ​​obtained by calculating dissimilarity or correlation indices. These values ​​reflect the strength and pattern of the correlation between different features. They can be a scalar value, a vector, or a matrix, representing the interdependence between features. Environmental impact parameters are the specific ways and intensity of the effects of aquatic environmental factors on the visual characteristics of diseases, determined by analyzing feature association metrics. These can be a single value or a set of values, such as brightness offset, color correction factor, texture blur coefficient, or contrast adjustment ratio, used to guide subsequent image processing.

[0100] To more accurately analyze the correlation patterns between aquatic environmental factors and image features of diseased areas, this method, when deriving environmental correlation features based on the visual representation features of the aquatic environment and the visual manifestation features of the disease, first performs feature dimension matching between the visual representation features of the aquatic environment and the visual manifestation features of the disease, thus obtaining matched feature pairs. This matching process ensures that features from different sources can be meaningfully compared at the semantic or visual level, avoiding invalid cross-dimensional analysis. For example, the influence of water turbidity on light scattering can be aligned with the brightness changes in the diseased area, or the color deviation of the water body can be aligned with the chromaticity anomaly of the diseased area. Based on this, the difference or correlation index between the matched feature pairs is calculated, thus obtaining a feature correlation metric. This calculation step transforms qualitative correlation into quantitative numerical values. For example, by calculating Euclidean distance or Pearson correlation coefficient, the correlation strength and pattern between aquatic environmental features and the visual manifestation features of the disease can be quantified, thereby revealing the specific degree of correlation between increased water turbidity and decreased contrast in the diseased area. Finally, based on the obtained feature correlation metrics, the degree and direction of influence of aquatic environmental factors on the visual characteristics of diseases are determined, and environmental impact parameters are obtained and used as environmental correlation features. This core step further analyzes the quantified correlation degree into actionable parameters, such as clarifying the specific degree and direction of the reduction in brightness in the diseased area caused by water turbidity. Through this refined matching, measurement, and analysis process, this method can more accurately quantify the specific impact of aquatic environmental factors on the visual characteristics of diseases, thereby obtaining environmental correlation features with higher precision. These high-precision environmental correlation features, as support for subsequent identification of the visual patterns of diseases presented in the aquatic environment, can directly guide the adaptive adjustment of local transformation parameters in style transfer operations, ensuring that disease features are preserved and even enhanced while removing environmental interference. This avoids the problem of disease information loss or distortion that may occur in traditional style transfer, thereby improving the accuracy of aquatic disease identification and early warning capabilities.

[0101] In some embodiments, the step of identifying visual patterns of disease characteristics in an aquatic environment and obtaining environmental disease pattern information by combining disease area, protection intensity parameters, and environmental correlation characteristics includes:

[0102] Pixel value distribution analysis, local gradient calculation, and color space conversion are performed on the image data of the diseased area to obtain the original visual feature parameters of the diseased area.

[0103] Based on environmental correlation characteristics, the original visual feature parameters of the diseased area are corrected for environmental impact to obtain the corrected visual representation of the diseased features in the current water environment.

[0104] By combining protection strength parameters, the corrected visual manifestations of disease characteristics in the current water environment are described or quantified in a patterned manner to obtain environmental disease pattern information.

[0105] Pixel value distribution analysis refers to the statistical analysis of the brightness or color values ​​of pixels within an image region to reveal its overall statistical characteristics. This can be achieved by calculating pixel histograms, mean, variance, skewness, or kurtosis. Local gradient calculation involves capturing the intensity and direction of pixel value changes within an image region to reflect details such as edges and textures. This can be achieved using the Sobe I operator, Prewitt operator, Roberts operator, or Canny edge detection algorithm. Color space conversion involves transforming the color representation of an image from one model (e.g., RGB) to another (e.g., HSV, Lab, or YCbCr) to better separate brightness, chromaticity, or perceived uniformity. This can be achieved using methods such as RGB to HSV conversion, RGB to Lab conversion, or RGB to YCbCr conversion. Environmental impact correction refers to adjusting or correcting the original visual feature parameters based on environmental correlation characteristics to eliminate or mitigate the interference of environmental factors on the visual representation of diseases. This can be achieved using lookup table (LUT)-based correction, regression model-based correction, or image restoration algorithms based on physical models. Pattern description refers to transforming the corrected visual representation of disease into a pattern template or feature vector with structure or semantics. This can be achieved by defining a set of preset visual pattern templates, generating feature descriptors, or constructing feature embeddings based on deep learning. Quantization refers to transforming the corrected visual representation of disease into operable numerical parameters or discrete levels. This can be achieved by mapping visual features to normalized values, transforming them into discrete intensity levels, or generating weight parameters to control the algorithm.

[0106] This scheme aims to accurately identify the visual patterns of disease characteristics presented in specific aquatic environments, generating accurate environmental disease pattern information to provide more refined guidance for subsequent style transfer operations and ensure the integrity of disease characteristics. Specifically, firstly, by performing pixel value distribution analysis, local gradient calculation, and color space conversion on the image data of the diseased area, this scheme extracts the inherent, environmentally undisturbed visual attributes of the original image data of the diseased area, obtaining the original visual feature parameters of the diseased area. Pixel value distribution analysis reveals the statistical characteristics of the diseased area, such as brightness and contrast; local gradient calculation captures details such as edges and textures; and color space conversion helps to understand the color characteristics of the disease from different dimensions. These operations together construct a comprehensive description of the original visual characteristics of the diseased area, laying the foundation for subsequent environmental impact correction. Secondly, based on environmental correlation features, environmental impact correction is performed on the original visual feature parameters of the diseased area to obtain the corrected visual representation of the disease characteristics in the current aquatic environment. This scheme utilizes previously obtained environmental correlation features reflecting the correlation pattern between aquatic environmental factors and image features of the diseased area to correct the original visual characteristics of the diseased area. This correction effectively removes or compensates for the interference of environmental factors on the visual representation of diseases, making the corrected visual representation more realistically reflect the actual visual state of diseases in the current complex aquatic environment. Finally, combined with the protection intensity parameter, the corrected visual representation of disease characteristics in the current aquatic environment is described or quantified in a pattern-based manner, obtaining environmental disease pattern information. This scheme combines the environmentally corrected visual representation of diseases with preset protection intensity parameters. The protection intensity parameter indicates the priority or degree of protection for disease characteristics during style transfer. Through pattern-based description or quantification, this scheme transforms the corrected visual representation of diseases into actionable environmental disease pattern information. Overall, through the close coordination of the above three steps, this scheme achieves the removal of environmental interference from the original image data, accurately captures the true visual pattern of diseases in a specific aquatic environment, and transforms it into pattern information that can guide style transfer operations. This refined pattern recognition capability enables the subsequent generation of environmental adaptation protection strategies with local transformation parameters for style transfer operations to more accurately reflect the actual visual state of diseases in complex environments and make targeted adjustments based on the expected protection intensity. This significantly enhances the ability of style transfer operations to preserve the integrity of disease features, preventing disease information from being smoothed, blurred, or eliminated during the conversion process. As a result, it provides more reliable and discriminative image input for the final identification of aquatic organism diseases, improving the accuracy of disease identification and early warning capabilities.

[0107] In some embodiments, the step of overlaying style-transferred aquatic organism images with disease feature indicator maps to obtain an overlaid image includes:

[0108] The boundary information, disease type information, and disease severity of multiple disease areas can be extracted from the disease characteristic indicator map;

[0109] Based on the disease type information, determine the preset visual coding rules for different disease types; the preset visual coding rules include color, texture or highlight mode;

[0110] Adjust the visual intensity parameters of the preset visual coding rules according to the severity of the disease; the visual intensity parameters include color saturation, transparency, or the diffusion range of the highlighted area;

[0111] By combining the boundary information of the diseased area, the adjusted intensity parameters, and the preset visual coding rules, a local overlay rendering instruction is generated for each diseased area.

[0112] By using local overlay rendering commands, the style-transferred aquatic organism image is overlaid with the disease feature indicator map, and different disease areas are displayed differently to obtain the overlay display image.

[0113] Disease feature indication maps are graphic data generated by extracting disease features from aquatic organism images. These maps indicate the location, type, and severity of diseases within the image and can be implemented using pixel-level masks, vector graphics, or structured data tables. Preset visual encoding rules are style specifications for visual presentation of different disease types, and can be implemented using color maps, texture libraries, or rendering templates for highlighted areas. Visual intensity parameters are numerical values ​​used to control the intensity of the visual encoding rules, and can be implemented using color channel values, transparency percentages, or pixel expansion values ​​for highlighted areas. Local overlay rendering commands are specific image processing commands generated for specific disease areas in the image, combining their boundaries, type, severity information, and corresponding visual encoding and intensity adjustments. These commands can be implemented using graphics rendering API calls, pixel operation functions, or shader programs. Differentiated display refers to presenting diseased areas using different visual encoding and intensity parameters based on their different attributes (such as type and severity), so that each diseased area has distinguishable visual characteristics in the overlaid image. This can be achieved through methods such as layered rendering, regional coloring, or dynamic highlighting.

[0114] The overlay display method of this application aims to provide a more informative and interpretable method for visualizing diseases, thereby solving the problem that simple overlay displays cannot effectively distinguish and express complex disease information. This method first parses the boundary information, disease type information, and disease severity of multiple disease areas from the disease feature indicator map. This parsing process is the foundation for all subsequent differentiated displays, transforming the original indicator map into structured, operable data, clarifying the precise location, type, and extent of each lesion. Based on this, according to the parsed disease type information, the system determines preset visual coding rules for different disease types. These rules can include color, texture, or highlighting patterns, giving different types of diseases unique visual identifiers; for example, bacterial diseases can be assigned red, while parasitic diseases can be assigned blue texture. This typified visual coding allows observers to intuitively identify the various diseases present in the image. Furthermore, according to the disease severity, the system adjusts the visual intensity parameters of the preset visual coding rules. These parameters can include color saturation, transparency, or the diffusion range of the highlighted area. By mapping the severity of disease to visual intensity—for example, higher visual intensity for more severe diseases—the overlaid image not only displays the presence and type of disease but also intuitively reflects its development stage or degree of damage. This intensity adjustment mechanism provides a more refined assessment of disease status. Subsequently, combining the boundary information of the disease area, the adjusted intensity parameters, and preset visual coding rules, the system generates local overlay rendering instructions for each disease area. These instructions are customized, ensuring that each disease area can be visually presented according to its own characteristics. Finally, using these local overlay rendering instructions, the style-transferred aquatic organism image is overlaid with the disease feature indicator map, and different disease areas are displayed differently to obtain the overlay display image. By executing these local rendering instructions, the system no longer simply overlays two images but performs intelligent and selective rendering based on the specific attributes of each disease area. This differentiated display method enables the final overlay display image to clearly, accurately, and intuitively present the disease status of aquatic organisms, including the precise location, type, and severity of the disease. This overlay display method is closely integrated with other steps in the style transfer image recognition method for aquatic organisms in this application. By overlaying refined disease information onto the style-transferred aquatic organism images, subsequent disease recognition steps can obtain input images with more information and interpretability. This refined visualization not only improves the accuracy of the automated recognition model but also greatly enhances the efficiency and reliability of manual review, thereby effectively supporting early warning and precise intervention for aquatic organism diseases and solving the problem of confusion of information on multiple lesions and types of diseases in complex underwater environments.

[0115] In a specific embodiment, suppose the system acquires a style-transferred image of an aquatic organism, identifying two lesions: a severe bacterial infection and a mild parasitic infection. First, the system parses the boundary information, infection type (bacterial infection, parasitic infection), and severity (severe, mild) of these two lesion areas from the lesion feature indicator map. Next, based on the infection type information, the system determines preset visual encoding rules. For example, bacterial infections can be set to a red highlight mode, while parasitic infections can be set to a blue texture mode. Then, based on the severity of the infection, the system adjusts the visual intensity parameters of these preset visual encoding rules. For severe bacterial infections, the color saturation of the red highlight mode can be adjusted to a higher level, and the diffusion range of the highlighted area can be set to a larger extent. For mild parasitic infections, the transparency of the blue texture mode can be adjusted to a lower level, while maintaining texture detail. Subsequently, the system combines the boundary information of each lesion area, the adjusted intensity parameters, and the preset visual encoding rules to generate local overlay rendering instructions for the two lesion areas. These instructions specify how to apply highly saturated red highlights to areas of bacterial infection and how to apply a semi-transparent blue texture to areas of parasitic infection. Finally, using these local overlay rendering instructions, the system overlays the style-transferred aquatic organism image with the disease feature indicator map. During the overlay process, the system differentiates the display of different disease areas according to the instructions, resulting in a final overlay image where bacterial infection areas are highlighted with highly saturated red, while parasitic infection areas are displayed with a semi-transparent blue texture. Both are clearly distinguishable, and the severity is intuitively presented through visual intensity.

[0116] The overlay display method of this application extracts boundary information, disease type information, and disease severity from the disease feature indicator map. Based on this information, it determines differentiated visual coding rules and intensity parameters, and then generates local overlay rendering instructions for differentiated display. This effectively solves the problem of overly dense and confusing indicator maps caused by existing simple overlay display methods when aquatic organisms have multiple lesion areas or different types of diseases in complex underwater environments. This method enables aquaculture personnel to clearly distinguish different lesions or disease types, intuitively understand the precise location, type, and severity of the disease, and thus improve the precision of diagnosis. In addition, this refined and differentiated display method greatly improves the efficiency and accuracy of manual review and enhances users' trust in the system's identification results, providing more informative and interpretable visual support for the identification of aquatic organism diseases.

[0117] Reference Appendix Figure 2 This invention provides an image recognition system for style transfer in aquatic organisms, comprising:

[0118] The acquisition module 100 is used to acquire images of aquatic organisms in the aquaculture environment;

[0119] Extraction module 200 is used to extract disease features from the acquired aquatic organism images to obtain disease feature information;

[0120] The generation module 300 is used to generate a disease feature indication map based on disease feature information;

[0121] The style transfer module 400 is used to perform style transfer operations on the acquired aquatic organism images to obtain style-transferred aquatic organism images; wherein, the style transfer operation controls the image processing of the diseased area according to the disease feature information to preserve the integrity of the disease features.

[0122] The overlay module 500 is used to overlay the style-transferred aquatic organism image with the disease feature indicator map to obtain an overlay display image;

[0123] The image recognition module 600 is used to identify diseases in aquatic organisms by overlaying images.

[0124] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0125] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An image recognition method for style transfer in aquatic organisms, characterized in that, Includes the following steps: Acquire images of aquatic organisms in aquaculture environments; Disease feature extraction was performed on the acquired aquatic organism images to obtain disease feature information; Based on the disease characteristic information, generate a disease characteristic indicator map; A style transfer operation is performed on the acquired aquatic organism images to obtain style-transferred aquatic organism images. The style transfer operation controls the image processing of the diseased areas based on the disease feature information to preserve the integrity of the disease features. The style-transferred aquatic organism images are overlaid with disease feature indicator maps to obtain an overlaid display image; Disease identification of aquatic organisms is achieved by overlaying images; The steps for performing style transfer on acquired aquatic organism images to obtain style-transferred aquatic organism images include: Based on the disease characteristics information, determine the disease areas in the aquatic organism images that require disease feature protection and the protection intensity parameters; Local image feature analysis was performed on non-disease areas adjacent to diseased areas in aquatic organism images to obtain local image feature information of non-disease areas; Based on the local image features of the diseased area, protection intensity parameters, and non-diseased areas, the local transformation parameters of the style transfer operation are adaptively adjusted. Using adaptively adjusted local transformation parameters, a style transfer operation is performed to obtain style-transferred images of aquatic organisms. The steps for adaptively adjusting the local transformation parameters of the style transfer operation based on the diseased area, protection intensity parameters, and local image feature information of the non-diseased area include: Image feature analysis is performed on the boundary between diseased and non-diseased areas to identify the feature gradient or feature continuity at the boundary, and confidence information of the diseased area boundary is obtained by combining the disease feature information. Based on the local image feature information of non-disease areas, the influence of water environmental factors on image features is quantified to obtain environmental interference feature quantity; Based on the diseased area, protection strength parameters, confidence information of the diseased area boundary, and environmental interference features, local transformation rules are generated for different image regions. The local transformation parameters of the style transfer operation are adjusted using local transformation rules.

2. The image recognition method for style transfer of aquatic organisms according to claim 1, characterized in that, The steps for extracting disease features from acquired aquatic organism images include: Identify local brightness, color, texture, or morphological changes in aquatic organism images that are related to disease.

3. The image recognition method for style transfer of aquatic organisms according to claim 1, characterized in that, The steps for adaptively adjusting the local transformation parameters of the style transfer operation based on the diseased area, protection intensity parameters, and local image feature information of the non-diseased area include: Based on the local image feature information of non-disease areas, the correlation pattern between water environmental factors and image features of disease areas is analyzed to obtain environmental correlation features; By combining the disease area, protection intensity parameters, and environmental correlation characteristics, the visual patterns of disease characteristics presented in the aquatic environment are identified, and environmental disease pattern information is obtained. Based on environmental disease pattern information, an environmental adaptation protection strategy is generated for local transformation parameters of style migration operations; By combining the local image features of the diseased area, the protection intensity parameters, and the non-diseased area, and by using an environmental adaptation protection strategy, the local transformation parameters of the style transfer operation are adjusted.

4. The image recognition method for style transfer of aquatic organisms according to claim 3, characterized in that, The steps for analyzing the correlation patterns between water environmental factors and image features of diseased areas based on local image feature information of non-disease-affected areas to obtain environmental correlation features include: Environmental factor characterization analysis was performed on local image feature information of non-disease areas to obtain visual characterization features of the water environment; Visual features of the disease are extracted from the image feature information of the diseased area to obtain the visual characteristics of the disease. Based on the visual characteristics of the aquatic environment and the visual manifestations of diseases, environmental correlation characteristics are obtained.

5. The image recognition method for style transfer of aquatic organisms according to claim 4, characterized in that, The steps to obtain environmental correlation characteristics based on the visual representation characteristics of the aquatic environment and the visual manifestation characteristics of diseases include: The visual representation features of the aquatic environment are matched with the visual manifestation features of diseases to obtain the matched feature pairs. Calculate the difference or correlation index between the matched feature pairs to obtain the feature association metric; Based on feature correlation measurement, the degree and direction of influence of aquatic environmental factors on the visual manifestation characteristics of diseases are determined, and environmental impact parameters are obtained and used as environmental correlation features.

6. The image recognition method for style transfer of aquatic organisms according to claim 4, characterized in that, The steps to identify visual patterns of disease characteristics in aquatic environments and obtain environmental disease pattern information by combining disease area, protection intensity parameters, and environmental correlation characteristics include: Pixel value distribution analysis, local gradient calculation, and color space conversion are performed on the image data of the diseased area to obtain the original visual feature parameters of the diseased area. Based on environmental correlation characteristics, the original visual feature parameters of the diseased area are corrected for environmental impact to obtain the corrected visual representation of the diseased features in the current water environment. By combining protection strength parameters, the corrected visual manifestations of disease characteristics in the current water environment are described or quantified in a patterned manner to obtain environmental disease pattern information.

7. The image recognition method for style transfer of aquatic organisms according to claim 1, characterized in that, The steps for overlaying style-transferred aquatic organism images with disease feature indicator maps to obtain the overlay image include: The boundary information, disease type information, and disease severity of multiple disease areas can be extracted from the disease characteristic indicator map; Based on the disease type information, determine the preset visual coding rules for different disease types; Adjust the visual intensity parameters of the preset visual coding rules according to the severity of the disease; By combining the boundary information of the diseased area, the adjusted intensity parameters, and the preset visual coding rules, a local overlay rendering instruction is generated for each diseased area. By using local overlay rendering commands, the style-transferred aquatic organism image is overlaid with the disease feature indicator map, and different disease areas are displayed differently to obtain the overlay display image.

Images