Product quality evaluation method and system based on deep neural network

Abstract

The invention discloses a product quality evaluation method and system based on a deep neural network, and belongs to the technical field of artificial intelligence. The method comprises the following steps: step 1, training a deep neural network into a quality evaluation model; 2, # imgabs0 # is input into an input layer of the quality evaluation model, and X is an evaluation index vector determined by an expert for a product; w is a coefficient vector of each evaluation index determined by an expert for a product, and an output layer of the quality evaluation model outputs an estimated value # imgabs1 #; 3, determining a first loss function; step 4, respectively adjusting each weight coefficient according to the first loss function; and step 5, obtaining a weight coefficient of each corresponding evaluation index when the first loss function is minimum. According to the method, the deep neural network is trained into the quality evaluation model, the quality of the product is further evaluated, the evaluation standard is unified, and subjectivity of different evaluators is avoided.

Description

Technical Field

[0001] The present invention relates to a product quality evaluation method and system based on a deep neural network, belonging to the field of artificial intelligence technology. Background Art

[0002] In the prior art, after constructing an evaluation index system for products, multiple experts were invited to assign weights to each layer of indicators in the index system. Through the analysis of the mean, dispersion degree, principal components, and comprehensive weights of the index weights, the importance of each index was initially analyzed. The index weight refers to the contribution degree of each index to the realization of the overall goal, and it reflects the coefficient of the value status of each index in the evaluation object. The weight directly affects the evaluation result, and the change of the weight value may cause the change of the order of the evaluation object's quality. In order to obtain good index weights, experts with rich experience, profound professional knowledge, different disciplinary backgrounds, and different work experiences in related fields were invited to evaluate the indicators. However, no matter how experienced the experts are, they are affected by their subjective understanding, and the weights assigned to each index are not objective. Summary of the Invention

[0003] To overcome the shortcomings of the prior art, the invention purpose of the present invention is to provide a product quality evaluation method and system based on a deep neural network, which can objectively assign weights to each index and make the evaluation result more objective.

[0004] To achieve the above invention purpose, the present invention provides a product quality evaluation method based on a deep neural network, which includes the following steps: Step 1: Train a deep neural network into a quality evaluation model for evaluating product quality; Step 2: Input into the input layer of the quality evaluation model, where is the vector of each evaluation index determined by an expert for the product, is the coefficient vector of each evaluation index determined by an expert for the product, and the output layer of the quality evaluation model outputs an estimated value , ; Step 3: Determine the first loss function according to the following formula: , where is the evaluation value of the product performance determined by an expert according to ; Step 4: Adjust each weight coefficient according to the first loss function L1: , where n = 1,..., N; is the learning rate; Step 5: Determine whether the first loss function L1 is the minimum. If it is, record the weight coefficients of each evaluation index corresponding to the minimum of the first loss function L1. If not, assign the changed weight coefficients to the original weight coefficients and return to Step 1.

[0005] To achieve the above-mentioned invention purpose, the present invention also provides a product quality evaluation system based on a deep neural network, which includes a storage medium and one or more processors. The storage medium stores a computer program, and the computer program is called by one or more processors to implement the above-mentioned product quality evaluation method based on a deep neural network.

[0006] Compared with the prior art, the present invention has the following beneficial effects:

[0007] By training a deep neural network into a quality evaluation model for evaluating product quality and fine-tuning the weight coefficients of the evaluation indexes, the present invention can objectively assign weights to each index, making the evaluation result more objective. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 is a flowchart of the product quality evaluation method based on a deep neural network provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0009] To make the technical means, creative features, achieved purposes and functions of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0010] In the present invention, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0011] Figure 1 is a flowchart of the product quality evaluation method based on a deep neural network provided by the present invention. As Figure 1 shown, the product quality evaluation method based on a deep neural network provided by the present invention includes the following steps: Step 1: Train a deep neural network into a quality evaluation model for evaluating product quality. The deep neural network includes a large language model; Step 2: Input into the input layer of the quality evaluation model, where is the vector of each evaluation index determined by an expert for the product, is the coefficient vector of each evaluation index determined by an expert for the product, and the output layer of the quality evaluation model outputs the estimated value , ; Step 3: Determine the first loss function according to the following formula: , where is the evaluation value of the product performance determined by an expert according to ; Step 4: Adjust each weight coefficient according to the first loss function L1: , where n = 1, …, N; is the learning rate; Step 5: Determine whether the first loss function L1 is the minimum. If so, record the weight coefficients of the corresponding evaluation indicators when the first loss function L1 is the minimum. If not, assign the changed weight coefficients to the original weight coefficients and return to Step 1.

[0012] In the present invention, the steps of training the deep neural network into a quality evaluation model include: S1-01: Obtain M groups of training data and M groups of test data, where M is an integer greater than or equal to 1; each group of training data includes , where , , i = 1, …, M; the test data is ; S1-02: Select K groups of data from the M groups of training data and M groups of test data, where K is a positive integer greater than or equal to 1 and less than or equal to M; S1-03: Input as a positive sample into the input layer of the deep neural network, and the output of the output layer of the deep neural network is , , is the function simulated by the neural network, is the current parameter of the deep neural network, k = 1, …, K; Input as a negative sample into the input layer of the deep neural network, and the output of the output layer of the deep neural network is , , , represents the gradient of ; is the adjustment coefficient; S1-04: Generate a second loss function according to and , and calculate the gradient of the parameter to be optimized of the deep neural network according to the second loss function ,

[0013] where is a hyperparameter; S1-05: Use the second loss function Calculate the gradient of the parameters of the deep neural network Update the current parameters of the deep neural network: , wherein, is the learning rate; S1-06: Determine whether the second loss function L2 is the minimum when training with K groups of data. If so, the training using K groups of data is completed, and then step S1-07 is executed; if not, assign the new parameters of the deep neural network to the original parameters and return to S1-03; S1-07: , and determine whether K is greater than M. If not, return to step S1-02; if so, end the training.

[0014] In the present invention, the deep neural network is trained into a quality evaluation model for evaluating product quality, and the weight coefficients of the evaluation indexes are finely adjusted, so that the weights of each index can be objectively assigned, making the evaluation result more objective. At the same time, the deep neural network is trained into a quality evaluation model through positive and negative samples, increasing the generalization ability of the quality evaluation model.

[0015] In the present invention, any index value is determined through the following steps: S2-01: Obtain Q samples of the nth index of the product, where Q is a positive integer greater than or equal to 10; S2-02: Assign Q to J, where J is an intermediate variable; S2-03: Calculate the average value of the nth index value of the product according to the following formula ; , wherein, is the qth sample of the nth index of the product; S2-04: Determine the suspicious samples according to the following formula : , wherein, is the sample set of the nth index; S2-05: Calculate the standard deviation of each sample according to the following formula : ; S2-06: Determine whether the suspicious sample is a singular value. If , is the adjustment coefficient, then If it is a singular value, filter it, then assign J - 1 to J, and return to step S2 - 03; otherwise, execute step S2 - 07; S2 - 07: Calculate the distribution function of the nth index of the product based on the D samples of the nth index of the product from which the singular values have been removed , d = 1, 2, …, D; S2 - 08: Generate A samples according to the distribution function: , a = 1, 2, …, A, where A is a positive integer greater than or equal to Q - D; S2 - 09: Calculate the value of the nth index of the product according to the following formula: , In the formula, is a dynamic adjustment coefficient that changes with time t.

[0016] Through the above technical solutions, the present invention can achieve the following beneficial effects: By constructing a distribution function from the limited test samples from which the singular values have been removed, generating data according to the distribution function to supplement the test samples, and then estimating the performance indicators of the product, the test cost is greatly reduced and the efficiency is improved.

[0017] Optionally, any index value is determined through the following steps: S3 - 01: Obtain Q samples of the nth index of the product, where Q is a positive integer greater than or equal to 10; S3 - 02: Calculate the average value of the nth index value of the product according to the following formula : , in the formula, is the qth sample of the nth index of the product; S3 - 03: Calculate the standard deviation of each sample according to the following formula : ; S3 - 04: Calculate the distribution function of the nth index of the product based on the q samples of the nth index , j = 1, 2, …, J; S3 - 05: Generate B samples according to the distribution function: , b = 1, 2, …, B, where B is a positive integer greater than or equal to Q; S3 - 06: Calculate the value of the nth index of the product according to the following formula .

[0018] The present invention can achieve the following beneficial effects through the above technical solutions: By constructing a distribution function from limited test samples, generating data according to the distribution function to supplement the test samples, and then estimating the performance indicators of the product, the test cost is greatly reduced and the efficiency is improved.

[0019] Optionally, any index value can also be determined through the following steps: S4-01: Establish a mixture model, which includes N distributions, where N is the number of indicators for evaluating quality; S4-02: Obtain C test samples of the product, and according to the following formula, attribute the c-th test sample to the data of the n-th quality evaluation index, c = 1,..., C, and C is a positive integer greater than or equal to 10: , In the formula, is the probability that the c-th test sample belongs to the component of the n-th distribution, U is the set of evaluation indicators, ; is the mixing coefficient; is the mean and standard deviation of the n-th distribution; S4-03: Generate multiple data according to the N distributions respectively to supplement the data of each evaluation index, so that the data under the N evaluation indexes all include B data, thus forming a data matrix with B rows and N columns; S4-04: Calculate the n-th index value of the product according to the following formula: .

[0020] The present invention can achieve the following beneficial effects through the above technical solutions: By first classifying the data, then constructing a distribution function from limited test samples, generating data according to the distribution function to supplement the test samples, and then estimating the performance indicators of the product, the test cost is greatly reduced and the efficiency is improved.

[0021] The present invention also provides a product quality evaluation system based on a deep neural network, which includes a storage medium and one or more processors. The storage medium stores a computer program, and the computer program is called by one or more processors to implement the above product quality evaluation method based on a deep neural network.

[0022] The present invention also provides a computer program product, which compiles the above product quality evaluation method based on a deep neural network into a computer program called and executed by one or more processors using a computer language.

[0023] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and what is described in the above embodiments and the specification is only to illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A product quality evaluation method based on deep neural network, characterized in that: The steps include: Step 1: Train the deep neural network into a quality assessment model to evaluate product quality; Step 2: Input to the input layer of the quality assessment model, where are the evaluation index vectors determined by experts for the product. is the coefficient vector of each evaluation index determined by experts for the product. The output layer of the quality evaluation model outputs the estimated value of product performance. , , Step 3: Determine the first loss function according to the following formula: , where Experts based Determine the estimated value of the product performance; Step 4: Adjust each weight coefficient according to the first loss function L1: , where n=1,…,N; is the learning rate; Step 5: Determine whether the first loss function L1 is the minimum. If so, record the weight coefficients of the corresponding evaluation indicators when the first loss function L1 is the minimum. If not, assign the changed weight coefficient to the original weight coefficient and return to step 1.

2. The product quality method based on deep neural network according to claim 1 is characterized in that: The steps to train a deep neural network into a quality assessment model include: S1-01: Obtain M groups of training data and M groups of test data, where M is an integer greater than or equal to 1; each group of training data includes ,in , , i=1,…,M; the test data is ; S1-02: Select K groups of data from M groups of training data and M groups of test data, where K is a positive integer greater than or equal to 1 and less than or equal to M; S1-03: As a positive sample input to the input layer of the deep neural network, the output of the deep neural network output layer is , , is the function simulated by the deep neural network, is the current parameter of the deep neural network, k=1,…,K; Will As a negative sample input to the input layer of the deep neural network, the output of the output layer of the deep neural network is , , , Express The gradient of The adjustment factor S1-04: According to and Generate the second loss function , according to the second loss function Calculate the parameters to be optimized for deep neural networks Gradient ; S1-05: Using the second loss function Gradients with respect to the parameters of a deep neural network Update the current parameters of the deep neural network: ， In the formula, is the learning rate; S1-06: Determine whether the second loss function L2 is the minimum when training with K groups of data. If so, complete the training with K groups of data, and then execute step S1-07; if not, assign new parameters of the deep neural network to the original parameters, and return to S1-03; S1-07: , and determine whether K is greater than M. If not, return to step S1-02, if yes, end the training.

3. The product quality evaluation method based on deep neural network according to claim 2 is characterized in that: Any indicator value Determine by following these steps: S2-01: Obtain Q samples of the nth indicator of the product, where Q is a positive integer greater than or equal to 10; S2-02: Assign Q to J, where J is an intermediate variable; S2-03: Calculate the average value of the nth indicator value of the product according to the following formula : , where is the qth sample of the nth indicator of the product; S2-04: Determine suspicious samples according to the following formula : ， In the formula, is the sample set of the nth indicator; S2-05: Calculate the standard deviation of each sample according to the following formula : ; S2-06: Identify suspicious samples Is it a singular value? If , is the adjustment factor, then For singular values, filter, and then assign J-1 to J, return to step S2-03; otherwise, execute step S2-07; S2-07: Calculate the distribution function of the product's nth indicator based on D samples of the product's nth indicator with singular values ​​removed , d=1,2,…,D; S2-08: Generate A samples according to the distribution function: ,a=1,2,…,A,A is a positive integer greater than or equal to QD; S2-09: Calculate the nth index value of the product according to the following formula: ， In the formula, is the dynamic adjustment coefficient over time t.

4. The product quality evaluation method based on deep neural network according to claim 2 is characterized in that: Any indicator value Determine by following these steps: S3-01: Obtain Q samples of the nth indicator of the product, where Q is a positive integer greater than or equal to 10; S3-02: Calculate the average value of the nth indicator value of the product according to the following formula : , where is the qth sample of the nth indicator of the product; S3-03: Calculate the standard deviation of each sample according to the following formula : ; S3-04: Calculate the distribution function of the nth indicator of the product based on q samples of the nth indicator , j=1,2,…,J; S3-05: Generate B samples according to the distribution function: ,b=1,2,…,B,B is a positive integer greater than or equal to Q; S3-06: Calculate the nth index value of the product according to the following formula: 。 5. The product quality evaluation method based on deep neural network according to claim 2 is characterized in that: Any indicator value Determine by following these steps: S4-01: Establish a hybrid model, wherein the hybrid model includes N distributions, where N is the number of indicators for evaluating quality; S4-02: Get C test samples of the product and convert the cth test sample into Data belonging to the nth evaluation quality indicator, c=1,…,C, C is a positive integer greater than or equal to 10: ， In the formula, is the cth test sample The probability of a component belonging to the nth distribution, U is a set of evaluation indicators, ; is the mixing coefficient; is the mean and standard deviation of the nth distribution; S4-03: Generate multiple data according to the N distributions to supplement the data of each evaluation index, so that the data under the N evaluation indexes include B data, thereby forming a data matrix with B rows and N columns; S4-04: Calculate the nth index value of the product according to the following formula: 。 6. A product quality evaluation system based on deep neural network, characterized in that: It includes a storage medium and one or more processors, wherein the storage medium stores a computer program, and the computer program is called by one or more processors to implement the product quality evaluation method based on deep neural network as described in any one of claims 1-5.

Images