CN-121981978-A - Automatic evaluation method for fat deposition state of fish liver

CN121981978ACN 121981978 ACN121981978 ACN 121981978ACN-121981978-A

Abstract

The invention relates to the technical field of fish detection, in particular to an automatic evaluation method of a liver fat deposition state of fish based on image segmentation and color feature analysis, which comprises the following steps of image acquisition, organ segmentation, feature extraction, color moment features of H, S, V channels are calculated by extracting organ images based on segmentation masks of organ areas and converting the organ images into HSV color space, fat deposition judgment, namely the color moment features are input into a classification model to obtain a fat deposition state evaluation result of liver, and the target organ areas can be automatically identified in an anatomical image, and the liver fat deposition state is objectively and quantitatively evaluated based on the color features with definite physical significance. By constructing an image processing flow consisting of organ segmentation, color feature extraction and classification judgment, the invention can realize efficient, standardized and interpretable automatic assessment of the fat deposition state of the fish liver.

Inventors

WANG ZHEN
ZHANG HUANLE
CUI YIFEI
WU YANG
Fu Hongyou
ZHANG LU
DONG SHAOGUANG
HU XIN

Assignees

国信工船(青岛)海洋科技有限公司
山东大学
青岛国信蓝色硅谷发展有限责任公司

Dates

Publication Date: 20260505
Application Date: 20260111

Claims (10)

1. An automated fish liver fat deposition state assessment method is characterized by comprising the following steps: S1, acquiring an image, namely acquiring a fish anatomy image; S2, organ segmentation, namely inputting the anatomical image into an image segmentation model, and segmenting liver, spleen and gonad areas to obtain corresponding segmentation masks; S3, extracting features, namely extracting an organ image based on a segmentation mask of the organ region, converting the organ image into an HSV color space, and calculating color moment features of each channel of H, S, V, wherein the color moment at least comprises a first moment, a second moment and a third moment; S4, fat deposition judgment, namely inputting the color moment characteristics into a classification model, and executing fat deposition grade judgment based on the color moment characteristics to obtain a fat deposition state evaluation result of the liver.
2. The automated fish liver fat deposition state assessment method of claim 1, wherein the image segmentation step comprises identifying liver, spleen and gonad regions based on a convolutional network, an attention mechanism or a prompt-driven segmentation model, and obtaining corresponding region masks, wherein the structure and weight of the segmentation model are not limited, and the segmentation model is within a range of realizable semantic segmentation models known in the art.
3. The automated fish liver fat deposition state assessment method of claim 2, wherein the segmented liver region mask is subjected to noise removal, hole filling and edge smoothing to obtain a clean liver region for subsequent color analysis.
4. The automated fish liver fat deposition state assessment method of claim 1, wherein the liver region is converted from RGB space to HSV space, wherein HSV space is used to enhance fat deposition-related color differential expression capability.
5. The automated fish liver fat deposition state assessment method of claim 1, wherein the color features comprise at least any one of the following statistical moment features calculated based on H, S, V channels: average, first-order central trend quantity; Variance or discrete measure; skewness or color distribution skewness measure.
6. The automated fish liver fat deposition state assessment method of claim 5, wherein the color features constitute at least 9-dimensional or more color statistics moment feature vectors and are used to reflect the regularity of the color distribution of liver regions.
7. The automated fish liver fat deposition state assessment method of claim 1, wherein the fat deposition state determination comprises inputting a preset classification model based on the color feature vector, classifying the liver sample in at least two or more classes, the classification result comprising at least one of normal, mild, moderate, or severe fat deposition.
8. The automated fish liver fat deposition state assessment method of claim 7, wherein the classification model is an interpretive model including, but not limited to, decision trees, logistic regression, K-nearest neighbor, or support vector machines.
9. The automated fish liver fat deposition state assessment method of claim 1, wherein the classification model determines liver fat deposition levels based on a threshold, interval division, or feature combination relationship of at least one statistical moment feature in a color feature vector.
10. The automated fish liver fat deposition state assessment method of claim 2, wherein the image segmentation model is a prompt-driven segmentation model, and the liver region mask is generated based on user interaction points, boundary prompts, or text prompts.

Description

Automatic evaluation method for fat deposition state of fish liver Technical Field The invention relates to the technical field of fish detection, in particular to an automatic evaluation method of a fish liver fat deposition state based on image segmentation and color feature analysis. Background With the large-scale and fine development of the aquaculture industry, the assessment of the liver condition of fish becomes a core link in the culture management. Liver is used as an important organ for metabolism and detoxification in fish, and the fat deposition degree is closely related to the nutrition level, metabolic capacity and health state of feed, so that the liver is an important index of fish growth and disease risk. Traditional liver fat deposition assessment methods mainly rely on manual dissection and visual judgment, and experience grading is carried out according to the color, glossiness and texture of livers and auxiliary organs. The method has obvious limitations that the evaluation process depends on manual experience, the subjectivity is strong, the judgment standards among different people are inconsistent, quantifiable indexes are difficult to form, the efficiency is low in large-scale sample processing, and the requirements of modern aquaculture on standardization, automation and batch detection are difficult to meet. In recent years, with the development of image processing and computer vision techniques, detection of a fish tissue state by an image analysis method has been attracting attention. However, the existing researches are mainly focused on disease identification, body surface damage detection or overall physiological state analysis, and the automatic evaluation of the viscera tissues of the fishes still has obvious defects. Specifically: (1) Organ-level image segmentation is limited. Many traditional methods rely on manual labeling of regions or simple segmentation algorithms based on threshold values and color differences, and are difficult to accurately distinguish tissue regions with similar morphologies and colors such as livers, spleens and gonads, and in an anatomical scene, organ boundaries are often blurred due to interference of factors such as blood, illumination and shielding, and the traditional method has low segmentation accuracy. (2) There is a lack of feature extraction methods for liver fat deposition. The existing image analysis method is mostly dependent on deep convolution characteristics to perform end-to-end learning, but the characteristics have poor interpretability, are sensitive to training data quantity, are difficult to directly establish a corresponding relation with physiological characteristics of fat deposition degree, and meanwhile, the deep learning model has higher calculation resource consumption and is not suitable for being deployed in an actual production environment. For the index of liver fat deposition, which is directly related to color change, the existing method lacks a stable, lightweight and clearly physically significant image feature extraction mechanism. (3) The automated evaluation system is not yet complete. Part of researches try to evaluate the liver fat state by deep learning, but often rely on end-to-end training, require a large amount of labeled data, and lack complete pipeline design of organ segmentation, color feature analysis and classification models, so that accuracy, interpretability and deployment cost cannot be considered. In summary, there is no method for accurately dividing a fish organ region in an anatomical image, and performing light-weight and interpretable automatic assessment on a liver fat deposition state based on color features with clear physiological significance in the prior art. Therefore, a new technical scheme is necessary to provide, so that the combination of organ-level image segmentation and color moment feature extraction is realized, and the accurate, stable and objective automatic evaluation of the fat deposition state of the fish liver is performed. Disclosure of Invention In order to solve the problems that in the prior art, fish liver fat deposition evaluation depends on manual observation, has strong subjectivity, is difficult to realize organ-level automatic segmentation, unstable in color feature extraction, insufficient in automation degree and the like, the invention provides a method capable of automatically identifying a target organ region in an anatomical image and objectively and quantitatively evaluating the liver fat deposition state based on color features with definite physical significance. By constructing an image processing flow consisting of organ segmentation, color feature extraction and classification judgment, the invention can realize efficient, standardized and interpretable automatic assessment of the fat deposition state of the fish liver. In order to achieve the above purpose, the present invention provides the following technical solutions: An automated ass