A fabric wrinkle rating method based on an improved AVO algorithm and optimized RRVFL

Abstract

This invention discloses an objective rating method for fabric wrinkles based on an improved AVO algorithm-optimized RRVFL. First, datasets of wrinkles from different types of fabrics are acquired and preprocessed. Then, the random weights, biases, and regularization coefficients of the RRVFL network are optimized using an improved AVO algorithm, and the output weights are calculated to obtain a trained HHO-AVO-RRVFL model. Finally, images of the fabric wrinkles to be rated are acquired and input into the trained HHO-AVO-RRVFL model to obtain fabric wrinkle rating information. This invention can detect and rate different types of wrinkles appearing in textiles during the textile industry production process, improving detection efficiency, thereby increasing production efficiency and improving product quality.

Description

Technical Field

[0001] This invention belongs to the field of objective fabric wrinkle rating technology, specifically relating to an objective fabric wrinkle rating method based on an improved AVO algorithm to optimize RRVFL. Background Technology

[0002] The overall quality of a fabric depends on many factors, including strength, shrinkage, abrasion resistance, fabric structure, appearance, wear, and wrinkling. Fabric wrinkles are a common phenomenon in daily life, unavoidable due to the characteristics of the fabric itself and the complexity of the environment in which it is used. However, wrinkles usually affect the appearance and comfort of clothing. Besides impacting aesthetics and comfort, wrinkled areas of fabric are more prone to wear and tear than smooth areas, thus reducing the fabric's lifespan. Therefore, wrinkle resistance is an important standard for evaluating fabric performance, becoming a crucial factor to consider when choosing fabrics or clothing. Manufacturers analyze the wrinkle resistance of fabrics before producing clothing and other goods. Studying wrinkle resistance requires comparing the wrinkle situation before and after wrinkling, that is, making a reasonable classification of the wrinkle level before and after wrinkling.

[0003] Since the early 1950s, many methods for assessing wrinkling have been developed. One of the most widely used methods is the American AATCC test, which allows expert observers to compare fabric samples to an AATCC replica standard and then score the fabric based on the similarity. The replica of the wrinkles illustrates five levels of wrinkles at one end, with level 1 representing a deeply wrinkled surface, while at the other end, level 5 represents a relatively smooth, intact surface.

[0004] After classifying fabric wrinkles into multiple levels, professionals classify these levels visually. However, this traditional method of visual judgment and classification is ineffective because human emotions, ambient lighting, and visual fatigue from prolonged wrinkle differentiation all affect accuracy, making the classification lack objectivity. Furthermore, analyzing a large volume of fabric wrinkles is time-consuming and labor-intensive, leading to a continuous decline in rating accuracy. Therefore, it is necessary to address these problems by using equipment to acquire fabric images and then having computers extract features from those images for rating. Summary of the Invention

[0005] The purpose of this invention is to address the problems of low accuracy and slow speed in existing objective rating methods for fabric wrinkles. It proposes an AVO algorithm (African vulturesoptimization) based on HHO (Harris Hawks Optimization, HHO) to optimize the parameters of the Random Vector Functional Link Network (RVFL), and uses the optimized RVFL to perform objective rating of fabric wrinkles.

[0006] This invention provides an objective rating method for fabric wrinkles based on an improved AVO algorithm to optimize RRVFL. The method includes the following steps:

[0007] Step 1: Obtain fabric images with wrinkle rating labels and preprocess them to form a dataset;

[0008] Step 2: Set the initial population size as a and the population dimension as m×(n+1)+1 in the HHO algorithm;

[0009] In the AVO algorithm, the initial population size is set to a and the population dimension is set to m×(n+1)+1;

[0010] Set the maximum number of iterations, the number of input layer nodes n and the number of hidden layer nodes m in RRVFL;

[0011] Step (2.1) Randomly generate a set of row vectors, use the HHO algorithm to quantize and assign values ​​to the row vectors, and use the quantized and assigned row vectors as the search population of the AVO algorithm to improve the optimization and convergence effect of the AVO algorithm; the number of row vectors is equal to the population size of the HHO algorithm, and the dimension of the row vectors is equal to the population dimension of the HHO algorithm.

[0012] Step (2.2) Construct the objective function, calculate the fitness value of the search population based on the objective function, and select the search population with the best fitness value among all search populations as the best vulture, and select the search population with the second best fitness value as the second best vulture; other search populations move towards the best vulture and the second best vulture using the roulette wheel formula.

[0013] Step (2.3) repeats step (2.2), and each repetition of step (2.2) completes an update of all search populations once, and the number of repetitions is the maximum number of iterations;

[0014] Step (2.4) calculates the fitness values ​​of all search populations obtained in step (2.3) based on the objective function, and selects the search population with the best fitness value;

[0015] Step 3: The search population with the best fitness value obtained in step (2.4) is also in the form of a row vector. The first m×n elements of the search population are rearranged into an m×n matrix and used as the input weights of RRVFL. The m elements after the m×n elements in the search population are rearranged into an m×1 column vector and used as the hidden layer bias of RRVFL. The last element in the search population is used as the regularization coefficient C of RRVFL.

[0016] Step 4: Based on the input weights of RRVFL, the hidden layer biases of RRVFL, and the regularization coefficient C of RRVFL, the output weights β of RRVFL are calculated, and then the RRVFL model is obtained.

[0017] Use the dataset obtained in step 1 to train the RRVFL model and obtain the trained RRVFL model.

[0018] Step 5: Obtain images of the fabric wrinkles to be rated, input them into the trained RRVFL model, and obtain the fabric wrinkle rating results.

[0019] Preferably, in step 3, the range of the element is between [-1, 1], and Abs(C×10) is used to map the regularization coefficient C to the interval (0, 10).

[0020] Preferably, in step 1, the fabric image includes photographs of various different fabrics, with the length and width of the photographs set to 224×224, and after regularization, mapped to a normal distribution function.

[0021] Preferably, in step 2, a is 40, the number of input layer nodes n is 224×224, the number of hidden layer nodes m is 80, and the maximum number of iterations is 60.

[0022] Preferably, in step 2, the objective function is the root mean square error function, the fitness value is the root mean square error value, and the fitness value with the minimum root mean square error value is the optimal fitness value.

[0023] The beneficial effects of this invention are as follows:

[0024] The purpose of this invention is to address the problems of low accuracy and slow speed in existing objective rating methods for fabric wrinkles. It proposes a method that uses a pre-trained deep learning neural network RVFL to extract image features and optimizes the parameters of RVFL using an AVO algorithm based on HHO. The optimized RVFL is then used to perform objective rating of fabric wrinkles. Attached Figure Description

[0025] Figure 1 This is a flowchart of the HHO-AVO-RRVFL algorithm constructed in this invention;

[0026] Figure 2 This is the RRVFL network structure in this invention;

[0027] Figure 3 This relates to the relationship between the number of hidden layer nodes and classification accuracy.

[0028] Figure 4 This relates the number of iterations and the error for different algorithm populations;

[0029] Figure 5 It refers to the classification performance of different algorithms;

[0030] Figure 6 These are box plots showing the stability of different algorithms;

[0031] Figure 5 and Figure 6 In the text, the correspondence between letters and models is as follows:

[0032] A: HHO-AVO-RRVFL; B: AVO-RRVFL; C: HHO-RRVFL; D: MFO-RRVFL; E: PSO-RRVFL; F: AOA-RRVFL; G: SSA-RRVFL; H: RRVFL. Detailed Implementation

[0033] The invention will now be further described with reference to the accompanying drawings.

[0034] like Figure 1 As shown, the purpose of this invention is to address the problems of low accuracy and slow speed in existing objective rating methods for fabric wrinkles. It proposes an AVO (African vultures optimization) algorithm based on HHO (Harris Hawks Optimization, HHO) to optimize the parameters of the Random Vector Functional Link Network (RVFL), and uses the optimized RVFL to perform objective rating of fabric wrinkles.

[0035] This invention includes the following steps:

[0036] Step 1:

[0037] (1.1) Obtaining the dataset

[0038] Industrial site data collection: To increase the diversity and usability of the dataset, some images were collected directly from textile industrial sites, including production lines and warehouses.

[0039] Web crawling: Using web crawling technology to retrieve relevant images from publicly available online resources, such as textile e-commerce platforms or professional forums.

[0040] Then, the data is filtered and organized.

[0041] Approximately 500 images were obtained through the above methods; these included fabrics mainly made of natural fibers (cotton, linen, silk, wool, etc.), synthetic fibers (polyester, nylon, acrylic, etc.), and blended materials. Wrinkle rating labels were added to the acquired images. (1.2) Prepare the dataset and perform preprocessing.

[0042] The images include photographs of various fabrics, all with a width and height set to 224×224. These images are then regularized and mapped to a normal distribution function. The dataset contains over 500 images. The dataset is divided into five folds for cross-validation, and the results are run 10 times, with the average value used as the final result.

[0043] Step 2:

[0044] Set the maximum number of iterations, the number of input layer nodes n and the number of hidden layer nodes m in RRVFL;

[0045] In the HHO algorithm, the initial population size is set to a and the population dimension is set to m×(n+1)+1;

[0046] In the AVO algorithm, the initial population size is set to a and the population dimension is set to m×(n+1)+1;

[0047] Step (2.1) Randomly generate a set of row vectors, use the HHO algorithm to quantize and assign values ​​to the row vectors, and use the quantized and assigned row vectors as the search population of the AVO algorithm to improve the optimization and convergence effect of the AVO algorithm; the number of row vectors is equal to the population size of the HHO algorithm, and the dimension of the row vectors is equal to the population dimension of the HHO algorithm.

[0048] Step (2.2) Construct the objective function, calculate the fitness value of the search population based on the objective function, and select the search population with the best fitness value among all search populations as the best vulture, and select the search population with the second best fitness value as the second best vulture; other search populations move towards the best vulture and the second best vulture using the roulette wheel formula.

[0049] Step (2.3) repeats step (2.2), and each repetition of step (2.2) completes an update of all search populations once, and the number of repetitions is the maximum number of iterations;

[0050] Step (2.4) calculates the fitness values ​​of all search populations obtained in step (2.3) based on the objective function, and selects the search population with the best fitness value;

[0051] Step 3:

[0052] The search population obtained in step (2.4) is also in the form of a row vector. The first m×n elements of the search population are rearranged into an m×n matrix and used as the input weights of RRVFL. The m elements after the m×n elements in the search population are rearranged into an m×1 column vector and used as the hidden layer bias of RRVFL. The last element in the search population is used as the regularization coefficient C of RRVFL.

[0053] Step 4: Train the RRVFL model optimized by the HHO-improved AVO algorithm.

[0054] The ultimate goal of this invention is to predict extracted features using the RVFL model. Therefore, it is necessary to first obtain an RVFL model optimized by the HHO-AVO algorithm. Because the input weights and hidden layer biases of RVFL are random, affecting the stability and accuracy of the model's classification results, the HHO-AVO algorithm is used to iteratively optimize these two parameters to obtain optimal parameter values, further improving the algorithm's accuracy. The algorithm model is as follows: Figure 1 As shown, the algorithm first extracts image features from the training set, and then obtains the optimal input weights, hidden layer bias, regularization coefficient C, and error combination through the obtained training set features HHO-AVO algorithm, further completing the training of the RVFL model, and obtaining the required output weights β of RVFL through ridge regression.

[0055] This step mainly explains how to obtain the optimal input weights, hidden layer bias, regularization coefficient C, and error through experiments, as well as the final output weights β of the RVFL, thereby deriving the trained model.

[0056] First, a brief introduction to the HHO algorithm, AVO algorithm, and RVFL network.

[0057] 1. HHO Algorithm

[0058] Harris Hawks Optimization (HHO) is a swarm intelligence optimization algorithm that mimics the hunting characteristics of Harris Hawks and combines Lévy Flights to solve complex, multidimensional problems. The HHO algorithm establishes a mathematical model of the Harris Hawk's hunting behavior; based on the escape energy of hares, the Harris Hawk will adopt corresponding encirclement strategies or search methods. It is similar to other metaheuristic algorithms.

[0059] The HHO algorithm includes two phases: exploration and development.

[0060] Each iteration updates the rabbit's location and fitness value. During the search phase, Harris Eagles employ two different search methods: one for rabbits found and another for rabbits not yet found. The rabbit's escape energy adaptively decreases with each iteration. When the rabbit's escape energy drops to a certain value, it will be spotted by the Harris Eagle flock, and the hunt transitions from the search phase to a raid phase. Before each raid, the rabbit has a certain probability of escaping the encirclement. Based on the rabbit's escape energy and whether it escapes the encirclement, Harris Eagles employ four different encirclement strategies. Each raid will ultimately result in the rabbit being captured, and each iteration will generate a new prey, whose location will be occupied by the rabbit with the highest fitness value in the new generation of the population.

[0061] The Harris Hawk can transition between different states based on the escape energy of the hare. Given the differences in escape energy among different hares, E0 (initial escape energy) is allowed to vary randomly within the range [-1, 1] during the algorithm iteration. The formula for escape energy is as follows:

[0062]

[0063] When the escape energy is greater than 1, the Harris Eagle is in the global reconnaissance phase. The Harris Eagle will either search for prey based on the location of other members and prey, or randomly perch on a tree within the population's range to detect prey.

[0064] Upon spotting prey, the Harris Eagle changes position based on the locations of other members and the prey. The formula is as follows:

[0065]

[0066] When the Harris Eagle has not found prey, it randomly perches within the population's range to search for prey.

[0067]

[0068] R(i) represents the position of the rabbit, r1 and r2 are random numbers, and ub and lb are the upper and lower bounds of the population position, respectively.

[0069] When the escape energy is less than 1, the Harris Eagle is in the development stage. The Harris Eagle selects one of four attack strategies based on the hare's escape energy and its own pursuit strategy. The hare's escape probability is a random value; less than 0.5 indicates a successful escape, and greater than 0.5 indicates a failed escape. The Harris Eagle typically uses either a fierce or gentle encirclement approach to capture its prey. This encirclement means the Harris Eagle will attack the prey gently or fiercely from different directions, depending on how much escape energy the prey has left. The Harris Eagle will get closer and closer to its target, increasing the chances of a successful kill through surprise attacks. After a period of time, the prey will lose more and more energy, at which point the Harris Eagle will intensify the encirclement to capture the prey. When the escape energy is greater than or equal to 0.5, a gentle encirclement is used; when the escape energy is less than 0.5, a fierce encirclement is used.

[0070] 2. AVO Algorithm

[0071] African Vultures Optimization (AVO) is a recently proposed algorithm based on the natural phenomenon of vultures migrating collectively to find food and coexist. The AVO algorithm provides a novel rotating search equation for optimization operations, employing vector coefficients F and r during the optimization process. A coefficient vector F with different random motions is created during the optimization operation, and this coefficient vector is used to modify the exploration and development phases. It utilizes suboptimal solutions in two different ways, leveraging Lévy-based flight patterns to improve local search through short, varied jumps. Two different mechanisms are used in the exploration phase to increase the diversity of the generated solutions, and four different mechanisms are provided in the development phase to enhance the algorithm's capabilities during the development phase. A brief description of the AVO algorithm is as follows:

[0072] Vultures can be physically divided into two groups. The algorithm first calculates the fitness function of all solutions (the initial population) to classify the vultures into categories. The optimal solution is the best first vulture, and the second solution is the second best vulture. Other vultures form a population that continuously approaches or replaces the optimal solution.

[0073] Similar to any other population-based metaheuristic algorithm, AVO begins with an initial set of random candidates. Therefore, the strength of each vulture is defined by applying it to the objective function under consideration. In this algorithm, the best vulture for each group is also identified and stored. A roulette wheel selection is used to determine the best candidate for each group.

[0074] The following formula describes the satiety rate of vultures:

[0075]

[0076]

[0077] In formula (2-9), the absolute value of F represents the vulture's satiety level. When the satiety level is greater than or equal to 1, the exploration phase begins. The vulture will randomly search for food in the surrounding area within a random distance of one of the two best populations.

[0078]

[0079] X(i)=X rand ×(1+b)-|(2×rand)×X rand -X(i)|×F(1+b) (7)

[0080] Or search randomly in the environment:

[0081]

[0082] When the satiety level is less than 1, the development phase begins. Development consists of two parts: if the satiety level is greater than or equal to 0.5, most of them will fight for food; if the satiety level is less than 0.5, the vultures will resort to violent siege tactics.

[0083] 3. RVFL network

[0084] Traditional Softmax classifiers suffer from high computational complexity, overfitting, sensitivity to outliers, and imbalanced classification. Unlike traditional classifiers, the weights of the hidden layer neurons in RVFL are randomly initialized, eliminating the need for backpropagation training. RVFL offers fast training speed, low computational complexity, and strong robustness against noise and overfitting, making it suitable for specific applications such as classification and regression tasks.

[0085] The following section describes the structure of the RVFL neural network. The input layer receives raw data as input, and the hidden layers contain a large number of linear neurons with random weights. In RVFL, the weights of the hidden layers are randomly initialized, and this randomness makes the network more robust to noise and overfitting. The network structure is as follows: Figure 2 As shown

[0086] The RVFL model can be represented as:

[0087]

[0088] Where X i =[x i1 ,x i2 ,......x in ] T ∈R n ;o i =[o i1 ,o i2 ,......,oim ] T ∈R m n and m are vector dimensions, and L is the number of hidden layer nodes. W j =[w j1 ,w j2 ,......,w jn ] T For input weights; b j =[b j1 ,b j2 ,......,b jn ] T β is the bias; g(x) is the activation function Sigmoid; d is the dimension of the input data; β j =[β j1 ,β j2 ,.......,β jm ] T For output weights; RVFL randomly generates W. j and b j .

[0089] Formula (5) can be simplified to:

[0090] Hβ=o (10)

[0091] In formula (6), matrix H is obtained by multiplying the input by the input weights and adding the hidden layer bias, and then activating it with the activation function. Matrix H is obtained by concatenating the input matrix H2 with the input matrix H2 through row operations; β is the output weight; o is the desired output; as shown in the following parts:

[0092]

[0093]

[0094]

[0095] In order to determine the output weights and minimize the output error To reach the minimum value of 0, let Therefore, equation (7) can be transformed into solving... Where H + It is the Moore-Penrose pseudo-inverse of matrix H.

[0096] 3. HHO-AVO-RRVFL Objective Rating Model for Fabric Wrinkles

[0097] Using HHO-AVO-RRVFL to rate fabric wrinkles demonstrates the high classification accuracy of HHO-AVO-RRVFL. The classification process of the HHO-AVO-RRVFL model is as follows:

[0098] HHO-AVO-RRVFL is used for feature classification. First, the dimensionality-reduced image features and label information are obtained. Then, the input weights (Weights), hidden layer bias (bias), and regularization coefficient (C) of RVFL are used as parameters for HHO-AVO. The position of each individual is updated according to the HHO-AVO algorithm, and the corresponding error is calculated. This error is used as the individual's fitness. If the calculated fitness is less than the current optimal fitness, the optimal solution in the population is updated; otherwise, no update is performed. The algorithm iterates until a maximum number of iterations is reached, at which point the position of the optimal solution is returned. The weights, bias, and C of RVFL are obtained by rearranging the matrix after the algorithm iterations to find the optimal solution. Next, the derivation of the output weight matrix β is performed. The input of RVFL is multiplied by the input weights (Weights) and the hidden layer bias (bias), activated by an activation function to obtain matrix H1. This matrix is ​​then concatenated with the input matrix H2 through row operations to form matrix H. The output weight matrix β is obtained by multiplying the Moore-Penrose pseudo-inverse of matrix H by the output matrix of RVFL.

[0099] Then, by combining the test set data with the previously obtained optimal Weights, bias, C, and derived β, the predicted values ​​for the test set are obtained. The HHO-AVO-RRVFL algorithm flow is as follows: Figure 1 As shown.

[0100] The following are the specific steps for model training.

[0101] (4.1) Selection of RVFL activation function

[0102] Different activation functions have different effects on RVFL classification. To select a relatively good activation function for RVFL, several different activation functions were selected for comparison. Table 1 shows the classification accuracy results of RVFL under these activation functions. The data in the table clearly shows that Sigmold is relatively better, and all subsequent experiments were conducted using Sigmold.

[0103] Table 1. Impact of activation function on classification accuracy

[0104]

[0105]

[0106] (4.2) The impact of the number of hidden layer nodes in RVFL on the algorithm

[0107] The number of hidden layer nodes in RVFL is also an important parameter. Too many nodes can lead to overfitting, while too few nodes result in low model accuracy. To find a relatively good value, the number of hidden layer nodes was set from 5 to 100, and the accuracy of different optimization algorithms was compared under different numbers. The experimental results are shown in Table 2.

[0108] Table 2. Impact of Number of Nodes on Classification Accuracy

[0109]

[0110] From Table 2 and Figure 3 It can be seen that the accuracy of different models increases with the number of samples, and tends to stabilize after 80. When the HHO-AVO optimization algorithm is selected, the accuracy is the highest. Based on the above experiments, in order to obtain a better accuracy, the number of samples is set to 80.

[0111] Step 5: Output fabric wrinkle rating results

[0112] After training the HHO-AVO-RRVFL model, the test set is input into the trained HHO-AVO-RRVFL model to obtain the final predicted rating of fabric wrinkles. Simultaneously, the test set is input into seven other trained benchmark models to obtain the corresponding classification results. Figures 3 to 4 The relationships between the number of hidden layer nodes and classification accuracy, and between the number of population iterations and error, are presented for eight benchmark models: AVO-RRVFL, HHO-RRVFL, MFO-RRVFL, PSO-RRVFL, AOA-RRVFL, SSA-RRVFL, and RRVFL, as well as the model HHO-AVO-RRVFL proposed in this invention.

[0113] Then, a comparative analysis of algorithm performance was conducted.

[0114] Eight algorithms were tested on the dataset. Each experiment was run 10 times, and the average value was taken to prevent errors. Table 3 shows the experimental results of different models.

[0115] Table 3 shows the experimental results on the dataset.

[0116]

[0117] Figure 5 This represents the ACC, Sensitivity, Specificity, and MCC metrics for different algorithms, along with the standard deviation of each metric. From... Figure 5It can be seen that HHO-AVO-RRVFL outperforms the other seven algorithms in all four evaluation metrics (ACC, sensitivity, specificity, and MCC). Regarding standard deviation, for the ACC metric, HHO-AVO-RRVFL's standard deviation is second only to HHO-RRVFL and PSO-RRVFL; for the sensitivity metric, it is second only to MFO-RRVFL; for the specificity metric, it is second only to AVO-RRVFL and HHO-RRVFL; and for the MCC metric, it is the lowest. Overall, HHO-AVO-RRVFL exhibits stronger and more stable classification accuracy than other algorithms, while its standard deviation is only slightly lower than HHO-RRVFL, demonstrating excellent stability in classification performance metrics. Analysis of the experimental results shows that the HHO-AVO-RRVFL model is the most effective model.

[0118] Algorithm stability analysis:

[0119] To verify the stability of the HHO-AVO-RRVFL algorithm, its box plots were compared with those of the other seven algorithms. Box plots were generated for the results of 10 iterations for each algorithm. Higher algorithm accuracy resulted in a higher median on the box plot. Algorithm stability was determined by comparing the size of the boxes, i.e., the distance between the upper and lower quartiles. Larger boxes with greater distances indicated less stability, and vice versa. The box plots were generated by... Figure 6 As shown, tests on the dataset revealed that the proposed HHO-AVO-RRVFL algorithm exhibits relatively high upper margins and upper quartiles compared to the other eight algorithms in its box plot. Only AVO-RRVFL and SSA-RRVFL have medians close to it, while its lower margins and lower quartiles are significantly higher than the other seven algorithms. The red plus signs indicate outliers; clearly, both MFO-RRVFL and SSA-RRVFL have data points that do not conform to the expected distribution of this algorithm. The box plots visually reflect the overall distribution and stability of each algorithm, with HHO-AVO-RRVFL showing the smallest box plot and significantly superior stability compared to the other algorithms.

[0120] Experiments show that the classification performance of the model proposed in this invention is generally superior to other algorithms, mainly in terms of maximum, minimum, and average accuracy. The model's stability is also better than most other algorithms. In conclusion, the model proposed in this invention possesses excellent classification prediction performance.

[0121] The above description is only a part of the specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A fabric wrinkle rating method based on an improved AVO algorithm to optimize RRVFL, characterized in that: The method includes the following steps: Step 1: Obtain fabric images with wrinkle rating labels and preprocess them to form a dataset; Step 2: Set the initial population size as a and the population dimension as m×(n+1)+1 in the HHO algorithm; In the AVO algorithm, the initial population size is set to a and the population dimension is set to m×(n+1)+1; Set the maximum number of iterations, the number of input layer nodes n and the number of hidden layer nodes m in RRVFL; Step (2.1) Randomly generate a set of row vectors, use the HHO algorithm to quantize and assign values ​​to the row vectors, and use the quantized and assigned row vectors as the search population of the AVO algorithm to improve the optimization and convergence effect of the AVO algorithm; the number of row vectors is equal to the population size of the HHO algorithm, and the dimension of the row vectors is equal to the population dimension of the HHO algorithm. Step (2.2) Construct the objective function, calculate the fitness value of the search population based on the objective function, and select the search population with the best fitness value among all search populations as the best vulture, and select the search population with the second best fitness value as the second best vulture; other search populations move towards the best vulture and the second best vulture using the roulette wheel formula. Step (2.3) repeats step (2.2), and each repetition of step (2.2) completes an update of all search populations once, and the number of repetitions is the maximum number of iterations; Step (2.4) calculates the fitness values ​​of all search populations obtained in step (2.3) based on the objective function, and selects the search population with the best fitness value; Step 3: The search population with the best fitness value obtained in step (2.4) is also in the form of a row vector. The first m×n elements of the search population are rearranged into an m×n matrix and used as the input weights of RRVFL. The m elements after the m×n elements in the search population are rearranged into an m×1 column vector and used as the hidden layer bias of RRVFL. The last element in the search population is used as the regularization coefficient C of RRVFL. Step 4: Based on the input weights of RRVFL, the hidden layer biases of RRVFL, and the regularization coefficient C of RRVFL, the output weights β of RRVFL are calculated, and then the RRVFL model is obtained. Use the dataset obtained in step 1 to train the RRVFL model and obtain the trained RRVFL model. Step 5: Obtain images of the fabric wrinkles to be rated, input them into the trained RRVFL model, and obtain the fabric wrinkle rating results.

2. The fabric wrinkle rating method based on the improved AVO algorithm to optimize RRVFL as described in claim 1, characterized in that, In step 3, the range of the element is between [-1, 1], and Abs(C×10) is used to map the regularization coefficient C to the interval (0, 10).

3. The fabric wrinkle rating method based on the improved AVO algorithm to optimize RRVFL as described in claim 1, characterized in that, In step 1, the fabric image includes photographs of various fabrics. The dimensions of the photographs are set to 224×224, and after regularization, they are mapped to a normal distribution function.

4. The fabric wrinkle rating method based on the improved AVO algorithm to optimize RRVFL as described in claim 1, characterized in that, In step 2, 'a' is 40, the number of input layer nodes 'n' is 224×224, the number of hidden layer nodes 'm' is 80, and the maximum number of iterations is 60.

5. The fabric wrinkle rating method based on the improved AVO algorithm to optimize RRVFL as described in claim 1, characterized in that, In step 2, the objective function is the root mean square error function, and the fitness value is the root mean square error value. The fitness value with the minimum root mean square error value is the optimal fitness value.

Images