Feather Classification Method Based on Fusion of Subjective and Objective Features
By integrating the subjective and objective characteristics of feather pieces and constructing a vectorized evaluation space using a residual neural network, the problems of poor classification consistency and low efficiency in existing technologies are solved, and efficient and accurate automatic classification of feather pieces is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ANHUI KEYI INTELLIGENT TECHNOLOGY CO LTD
- Filing Date
- 2026-02-02
- Publication Date
- 2026-06-02
AI Technical Summary
Existing technologies struggle to accurately and efficiently classify feathers by combining subjective and objective characteristics, and lack the quantification and transfer of professional judgment experience, resulting in poor classification consistency and low efficiency.
A classification method based on the fusion of subjective and objective features is adopted. By collecting and preprocessing feather images, combining the subjective feelings and ratings of professionals with objective physical parameters, a vectorized evaluation space is constructed using a residual neural network to learn complex evaluation criteria for classification.
It achieves highly accurate classification results that are highly consistent with those of professionals, significantly improving classification efficiency, reducing labor costs, and supporting customization of various evaluation criteria.
Smart Images
Figure CN122135074A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to feather classification, and more specifically to a feather classification method based on the fusion of subjective and objective characteristics. Background Technology
[0002] Feathers are the primary raw material for producing badminton shuttlecocks, and their quality directly determines the quality of the finished product. During shuttlecock production, feathers undergo rigorous classification and screening to ensure that different grades of shuttlecocks use feathers of corresponding quality. Traditional feather classification relies heavily on the subjective judgment of professionals, who classify feathers by observing their texture, luster, color, shape, and other characteristics, combined with tactile information. However, this manual classification method depends on subjective judgment, and different individuals may have different evaluation standards, leading to poor classification consistency, low efficiency, and high costs.
[0003] To address the aforementioned problems, existing technologies have proposed several methods for automatic feather classification based on image processing and machine learning. For example:
[0004] The invention patent application with publication number CN104697458A discloses a method for measuring the curvature of feathers. The method uses image processing technology to measure the curvature of feathers and obtains the curvature value through a calculation formula. This method mainly focuses on the measurement of physical parameters and does not involve the quantification of subjective evaluation factors.
[0005] The invention patent application with publication number CN105740883A discloses a method for classifying feather pieces based on contour and texture in a computer. This method extracts the LBP texture features and color features of the feather pieces and uses an SVM classifier for classification. Although the method extracts texture features, it extracts the original physical feature values of the texture (such as LBP histogram, color histogram, etc.) and does not convert these features into a subjective rating by professionals.
[0006] The aforementioned existing technologies have the following shortcomings:
[0007] First, existing technologies mainly extract the physical characteristics of feathers (such as curvature, texture, and color). There is a complex nonlinear relationship between these physical characteristics and the subjective evaluation of professionals. Simply classifying them based on these physical characteristics is difficult to accurately reflect the evaluation criteria of professionals.
[0008] Secondly, when evaluating feathers, professionals do not simply pursue a certain indicator that is "the higher the better" or "the lower the better," but rather seek an optimal state that is "just right." For example, in terms of texture, it is not that the finer or coarser the texture, the better, but rather that it should reach a moderate state; in terms of luster, it is not that the stronger or weaker the luster, the better, but rather that it should have a moderate luster; in terms of color, it is not that the whiter or yellower the better, but rather that it should meet specific standards for hue and uniformity. This complex non-monotonic relationship makes the relationship between physical characteristics and quality ratings present an "inverted U-shaped" or other complex non-linear curve relationship, which is difficult to accurately model using simple linear models or feature engineering.
[0009] Third, existing technologies lack effective means to quantify and transfer the subjective evaluation knowledge of professionals. The evaluation experience and standards accumulated by professionals cannot be transformed into replicable and transferable quantitative models. Summary of the Invention
[0010] (a) Technical problems to be solved
[0011] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a feather classification method based on the fusion of subjective and objective features, which can effectively overcome the shortcomings of the existing technology in that it is difficult to combine subjective and objective features to accurately and efficiently classify feathers.
[0012] (II) Technical Solution
[0013] To achieve the above objectives, the present invention provides the following technical solution:
[0014] The feather classification method based on the fusion of subjective and objective features includes the following steps:
[0015] S1. Acquire images of feathers and perform preprocessing;
[0016] S2. Collect feather images, and manually rate the texture fineness, luster, color and overall texture index. Use the feather images with rating labels to build a training dataset to train the subjective evaluation model.
[0017] S3. Input the image of the feather piece to be classified into the pre-trained subjective evaluation model. The subjective evaluation model outputs the subjective evaluation score of each indicator and constructs the subjective feeling rating vector.
[0018] S4. After measuring the length, width, and curvature of the feathers to be classified, normalize them and construct an objective feature vector.
[0019] S5. Construct a vectorized evaluation space, integrate subjective feeling rating vectors and objective feature vectors, map the feather pieces to be classified to specific locations in the vectorized evaluation space, and classify them according to their spatial locations.
[0020] Preferably, acquiring feather images in S1 includes:
[0021] Images of feathers are acquired using standardized image acquisition devices, which include:
[0022] Fixed LED light source to simulate natural light;
[0023] Standard black or yellow background board;
[0024] A fixed shooting angle, with the camera perpendicular to the plane of the feather.
[0025] Preferably, preprocessing is performed in S1, including:
[0026] Gaussian filtering for noise reduction;
[0027] Brightness equalization is achieved using histogram equalization.
[0028] Instance segmentation, trimming, and rotation of feather pieces eliminates random angle rotation during cutting.
[0029] Preferably, in S2, images of feather pieces are collected, and a person subjectively rates the texture fineness, luster, color, and overall texture index, including:
[0030] For each feather image, professionals subjectively rated its texture, luster, color, and overall quality.
[0031] Texture fineness: This is a subjective assessment of the texture's quality, rather than a numerical value for texture roughness. A high score indicates a fine, uniform, and orderly texture, showcasing high-quality characteristics.
[0032] Gloss perception: The subjective quality of gloss is judged, rather than the reflectance value. A high score indicates that the gloss is moderate, uniform, and natural, presenting a high-quality characteristic.
[0033] Color perception: Judging whether the color meets the standard, rather than the hue angle. A high score indicates that the color meets the standard, is uniform and consistent, and has no impurities.
[0034] Overall Texture: Evaluate the overall texture of the feather pieces;
[0035] Each indicator is worth 1 to 5 points.
[0036] Preferably, in S2, a training dataset is constructed using feather images with rating labels to train the subjective evaluation model, including:
[0037] Feather images with rating labels are input into a residual neural network ResNet for training. Based on the subjective rating data of professionals, it learns complex, non-monotonic evaluation criteria. Through multi-layer nonlinear transformation, it learns the evaluation patterns of professionals on various indicators, rather than simple linear positive or negative correlations.
[0038] Among them, the ResNet residual neural network supports customized training with multiple evaluation criteria, and the output subjective evaluation scores can be graded.
[0039] Preferably, in step S4, the length, width, and curvature of the feathers to be classified are measured and then normalized to construct an objective feature vector, including:
[0040] S41. Measure the length, width, and curvature of the feathers to be classified, extract the feather outline from the image of the feathers to be classified using image processing technology, detect the center line of the feather shaft, and use mathematical methods to measure and calculate various physical parameters;
[0041] S42. Since the data range of various physical parameters differs by several orders of magnitude, it is necessary to normalize each physical parameter, select a maximum measured value for each physical parameter, and calculate the ratio of each actual value to the maximum measured value.
[0042] S43. Construct an objective feature vector based on the ratio of each actual value to the maximum measured value;
[0043] The curvature is defined as the maximum deviation between the center line of the feather shaft and the horizontal line.
[0044] Preferably, in S5, a vectorized evaluation space is constructed, integrating subjective feeling rating vectors and objective feature vectors. The feather pieces to be classified are mapped to specific locations within the vectorized evaluation space, and classification is performed based on these spatial locations, including:
[0045] S51. Construct a vectorized evaluation space, integrate subjective feeling rating vector and objective feature vector to obtain a set of evaluation space coordinates, and map the feather pieces to be classified to the corresponding positions in the 8-dimensional vectorized evaluation space.
[0046] S52. In the vectorized evaluation space, classify all feathers by calculating distance;
[0047] In the vectorized evaluation space, feathers of similar quality are closer together, while feathers of different quality are farther apart.
[0048] Preferably, in S52, in the vectorized evaluation space, all feathers are classified by calculating distance, including:
[0049] First, determine the position of the manually selected standard feather pieces in the vectorized evaluation space, and then automatically determine the category based on the distance between the feather piece to be classified and the standard feather piece.
[0050] Preferably, after measuring the length, width, and curvature of the feathers to be classified in S4, performing normalization processing, and constructing an objective feature vector, the process includes:
[0051] Based on the pre-set classification thresholds corresponding to the values in the subjective feeling rating vector and the objective feature vector, the category of the feather piece to be classified is automatically determined.
[0052] (III) Beneficial Effects
[0053] Compared with existing technologies, the feather classification method based on the fusion of subjective and objective features provided by this invention uses deep learning to perform nonlinear mapping on original physical features such as texture, gloss, and color, transforming them into subjective ratings by professionals. It learns a complex, non-monotonic evaluation standard that is "just right," while simultaneously measuring objective physical parameters such as the length, width, and curvature of the feathers. By fusing subjective rating vectors and objective feature vectors, a vectorized evaluation space is constructed. This method is the first to quantify physical features into subjective ratings through nonlinear mapping, breaking through the limitations of simple linear relationships and solving the problem of difficulty in modeling complex evaluation standards for manual grading. It maintains a high degree of consistency with the classification results of professionals, has a significantly faster processing speed than manual classification, greatly reduces labor costs, supports the customization of multiple evaluation standards, and realizes the explicitness and solidification of professionals' tacit knowledge. Attached Figure Description
[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.
[0055] Figure 1 This is a schematic diagram of the process of the present invention;
[0056] Figure 2 This is a two-dimensional projection of the vectorized evaluation space in this invention;
[0057] Figure 3 This is a three-dimensional projection of the vectorized evaluation space in this invention. Detailed Implementation
[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0059] The following describes the specific process of the feather classification method based on the fusion of subjective and objective features provided by this invention, using a concrete example (e.g.) Figure 1 (as shown) and technical effects.
[0060] S1. Acquire images of feathers and perform preprocessing.
[0061] Specifically, S1 acquires images of feather segments, including:
[0062] Images of feathers are acquired using standardized image acquisition devices, which include:
[0063] Fixed LED light source to simulate natural light;
[0064] Standard black or yellow background board;
[0065] A fixed shooting angle, with the camera perpendicular to the plane of the feather.
[0066] Specifically, preprocessing is performed in S1, including:
[0067] Gaussian filtering for noise reduction;
[0068] Brightness equalization is achieved using histogram equalization.
[0069] Instance segmentation, trimming, and rotation of feather pieces eliminates random angle rotation during cutting.
[0070] S2. Collect images of feathers, and manually rate the texture, luster, color, and overall quality indicators. Use the feather images with rating labels to build a training dataset and train the subjective evaluation model.
[0071] Specifically, in S2, images of feather fragments are collected, and humans subjectively rate the texture fineness, luster, color, and overall texture indicators, including:
[0072] For each feather image, professionals subjectively rated its texture, luster, color, and overall quality.
[0073] Texture fineness: This is a subjective assessment of the texture's quality, rather than a numerical value for texture roughness. A high score indicates a fine, uniform, and orderly texture, showcasing high-quality characteristics.
[0074] Gloss perception: The subjective quality of gloss is judged, rather than the reflectance value. A high score indicates that the gloss is moderate, uniform, and natural, presenting a high-quality characteristic.
[0075] Color perception: Judging whether the color meets the standard, rather than the hue angle. A high score indicates that the color meets the standard, is uniform and consistent, and has no impurities.
[0076] Overall Texture: Evaluate the overall texture of the feather pieces;
[0077] Each indicator is worth 1 to 5 points.
[0078] Specifically, S2 uses feather images with rating labels to construct a training dataset to train the subjective evaluation model, including:
[0079] Feather images with rating labels are input into a residual neural network ResNet for training. Based on the subjective rating data of professionals, it learns complex, non-monotonic evaluation criteria. Through multi-layer nonlinear transformation, it learns the evaluation patterns of professionals on various indicators, rather than simple linear positive or negative correlations.
[0080] Among them, the ResNet residual neural network supports customized training with multiple evaluation criteria, and the output subjective evaluation scores can be graded.
[0081] S3. Input the image of the feather piece to be classified into the pre-trained subjective evaluation model. The subjective evaluation model outputs the subjective evaluation score of each indicator and constructs the subjective feeling rating vector.
[0082] S4. After measuring the length, width, and curvature of the feathers to be classified, normalize them and construct an objective feature vector, including:
[0083] S41. Measure the length, width, and curvature of the feathers to be classified, extract the feather outline from the image of the feathers to be classified using image processing technology, detect the center line of the feather shaft, and use mathematical methods to measure and calculate various physical parameters;
[0084] S42. Since the data range of various physical parameters differs by several orders of magnitude, it is necessary to normalize each physical parameter, select a maximum measured value for each physical parameter, and calculate the ratio of each actual value to the maximum measured value.
[0085] S43. Construct an objective feature vector based on the ratio of each actual value to the maximum measured value;
[0086] The curvature is defined as the maximum deviation between the center line of the feather shaft and the horizontal line.
[0087] S5. Construct a vectorized evaluation space, integrating subjective feeling rating vectors and objective feature vectors, mapping the feather pieces to be classified to specific locations in the vectorized evaluation space, and classifying them according to their spatial locations, such as... Figure 2 and Figure 3 As shown, it includes:
[0088] S51. Construct a vectorized evaluation space, integrate subjective feeling rating vector and objective feature vector to obtain a set of evaluation space coordinates, and map the feather pieces to be classified to the corresponding positions in the 8-dimensional vectorized evaluation space.
[0089] S52. In the vectorized evaluation space, classify all feathers by calculating distance;
[0090] In the vectorized evaluation space, feathers of similar quality are closer together, while feathers of different quality are farther apart.
[0091] In the technical solution of this application, the classification of feather fragments to be classified includes two methods: spatial location and preset threshold.
[0092] 1) Spatial location
[0093] In S52, all feathers are classified in the vectorized evaluation space by calculating distance, including:
[0094] First, determine the position of the manually selected standard feather pieces in the vectorized evaluation space, and then automatically determine the category based on the distance between the feather pieces to be classified and the standard feather pieces;
[0095] 2) Preset threshold
[0096] Based on the high interpretability of subjective rating vectors and objective feature vectors, after measuring the length, width, and curvature of the feathers to be classified in S4 and normalizing them, and constructing the objective feature vectors, the following techniques are directly applied:
[0097] Based on the pre-set classification thresholds corresponding to the values in the subjective feeling rating vector and the objective feature vector, the category of the feather piece to be classified is automatically determined.
[0098] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A feather classification method based on the fusion of subjective and objective features, characterized in that: Includes the following steps: S1. Acquire images of feathers and perform preprocessing; S2. Collect feather images, and manually rate the texture fineness, luster, color and overall texture index. Use the feather images with rating labels to build a training dataset to train the subjective evaluation model. S3. Input the image of the feather piece to be classified into the pre-trained subjective evaluation model. The subjective evaluation model outputs the subjective evaluation score of each indicator and constructs the subjective feeling rating vector. S4. After measuring the length, width, and curvature of the feathers to be classified, normalize them and construct an objective feature vector. S5. Construct a vectorized evaluation space, integrate subjective feeling rating vectors and objective feature vectors, map the feather pieces to be classified to specific locations in the vectorized evaluation space, and classify them according to their spatial locations.
2. The feather classification method based on the fusion of subjective and objective features according to claim 1, characterized in that: Images of feathers are acquired in S1, including: Images of feathers are acquired using standardized image acquisition devices, which include: Fixed LED light source to simulate natural light; Standard black or yellow background board; A fixed shooting angle, with the camera perpendicular to the plane of the feather.
3. The feather classification method based on the fusion of subjective and objective features according to claim 2, characterized in that: Preprocessing is performed in S1, including: Gaussian filtering for noise reduction; Brightness equalization is achieved using histogram equalization. Instance segmentation, trimming, and rotation of feather pieces eliminates random angle rotation during cutting.
4. The feather classification method based on the fusion of subjective and objective features according to claim 1, characterized in that: Images of feathers were collected in S2, and humans subjectively rated the texture, luster, color, and overall quality, including: For each feather image, professionals subjectively rated its texture, luster, color, and overall quality. Texture fineness: This is a subjective assessment of the texture's quality, rather than a numerical value for texture roughness. A high score indicates a fine, uniform, and orderly texture, showcasing high-quality characteristics. Gloss perception: The subjective quality of gloss is judged, rather than the reflectance value. A high score indicates that the gloss is moderate, uniform, and natural, presenting a high-quality characteristic. Color perception: Judging whether the color meets the standard, rather than the hue angle. A high score indicates that the color meets the standard, is uniform and consistent, and has no impurities. Overall Texture: Evaluate the overall texture of the feather pieces; Each indicator is worth 1 to 5 points.
5. The feather classification method based on the fusion of subjective and objective features according to claim 4, characterized in that: S2 uses feather images with rating labels to construct a training dataset to train the subjective evaluation model, including: Feather images with rating labels are input into a residual neural network ResNet for training. Based on the subjective rating data of professionals, it learns complex, non-monotonic evaluation criteria. Through multi-layer nonlinear transformation, it learns the evaluation patterns of professionals on various indicators, rather than simple linear positive or negative correlations. Among them, the ResNet residual neural network supports customized training with multiple evaluation criteria, and the output subjective evaluation scores can be graded.
6. The feather classification method based on the fusion of subjective and objective features according to claim 1, characterized in that: In S4, the length, width, and curvature of the feathers to be classified are measured, normalized, and an objective feature vector is constructed, including: S41. Measure the length, width, and curvature of the feathers to be classified, extract the feather outline from the image of the feathers to be classified using image processing technology, detect the center line of the feather shaft, and use mathematical methods to measure and calculate various physical parameters; S42. Since the data range of various physical parameters differs by several orders of magnitude, it is necessary to normalize each physical parameter, select a maximum measured value for each physical parameter, and calculate the ratio of each actual value to the maximum measured value. S43. Construct an objective feature vector based on the ratio of each actual value to the maximum measured value; The curvature is defined as the maximum deviation between the center line of the feather shaft and the horizontal line.
7. The feather classification method based on the fusion of subjective and objective features according to claim 1, characterized in that: In S5, a vectorized evaluation space is constructed, integrating subjective feeling rating vectors and objective feature vectors. The feather pieces to be classified are mapped to specific locations within this vectorized evaluation space, and classification is performed based on these spatial locations, including: S51. Construct a vectorized evaluation space, integrate subjective feeling rating vector and objective feature vector to obtain a set of evaluation space coordinates, and map the feather pieces to be classified to the corresponding positions in the 8-dimensional vectorized evaluation space. S52. In the vectorized evaluation space, classify all feathers by calculating distance; In the vectorized evaluation space, feathers of similar quality are closer together, while feathers of different quality are farther apart.
8. The feather classification method based on the fusion of subjective and objective features according to claim 7, characterized in that: In S52, all feathers are classified in the vectorized evaluation space by calculating distance, including: First, determine the position of the manually selected standard feather pieces in the vectorized evaluation space, and then automatically determine the category based on the distance between the feather piece to be classified and the standard feather piece.
9. The feather classification method based on the fusion of subjective and objective features according to claim 1, characterized in that: In S4, after measuring the length, width, and curvature of the feathers to be classified, normalization is performed, and an objective feature vector is constructed, including: Based on the pre-set classification thresholds corresponding to the values in the subjective feeling rating vector and the objective feature vector, the category of the feather piece to be classified is automatically determined.