Signal processing software quality prediction method based on neural network
By constructing a signal processing software quality model and using a BP neural network prediction method, the real-time and accuracy issues of signal processing software quality control were solved, improving the accuracy and reliability of quality prediction, reducing maintenance costs, and providing a systematic quality management solution.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SOUTHWEST CHINA RES INST OF ELECTRONICS EQUIP
- Filing Date
- 2025-12-30
- Publication Date
- 2026-04-10
AI Technical Summary
Existing signal processing software quality control relies on single-dimensional document and code reviews, lacking real-time performance and accuracy. Quality control is fragmented and difficult to achieve proactive prediction and accuracy. General quality models cannot adapt to complex nonlinear relationships, and there is a lack of effective quality prediction methods.
A quality model for signal processing software is constructed. A hierarchical framework combined with the GQM method is used to determine the metrics. A BP neural network is used to establish a prediction model for quality characteristics and sub-characteristics. The model is adjusted through a questionnaire survey. After the dataset is standardized, it is divided into training, validation and test sets. The parameters of the neural network model are optimized to ensure that the prediction accuracy reaches the set threshold.
It achieves structured characterization and standardization of signal processing software quality, improves prediction accuracy and reliability, enables proactive quality control, reduces maintenance costs, and provides a systematic quality management solution.
Smart Images
Figure CN121832893A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of software quality prediction, and particularly relates to a signal processing software quality prediction method based on a neural network. BACKGROUND
[0002] Signal processing software plays a vital role in modern technology and is widely used in communication, radar, navigation and other fields. With the rapid development of digital technology, signal processing software can efficiently collect, analyze, transform and identify various signals, thereby extracting useful information. Due to the sharp increase in data volume, the requirements for the accuracy and real-time performance of signal processing software quality prediction are increasingly high in order to make early software quality risk prevention and reduce maintenance costs. The existing quality prediction methods face the following challenges: 1) The historical data is not enough to guide the quality control of signal processing software, and the management decision is still mainly based on intuition. The existing quality control of signal processing software is mainly based on document, code review and testing, which brings the problems of single dimension of quality control data and insufficient real-time performance and accuracy, untimely quality control, and segmented quality control of each stage and each link. Quality analysis still relies on personal experience, and the generation, management and application of quality data lack means, and there is no connection with other dimension data of the product, and the experience and lessons formed by quality problems are difficult to effectively apply to prevent the recurrence of problems.
[0003] 2) Software quality control is mostly triggered by passive fault events, and the prediction accuracy of active prediction methods is not enough. Signal processing software is often the key to system cooperation and interaction, and the difficulty and cost of post-maintenance caused by quality problems are much higher than that in the early development stage. The focus of software quality control must be moved forward, i.e. from result-based analysis to process-based analysis. The current focus on moving forward often focuses on the summary of problems and experience to gradually improve the quality of requirements and design, and lacks more effective methods to prevent problems in advance.
[0004] 3) The general quality model and quality prediction method are not enough to guide the signal processing software. Signal processing software is faced with the development status of large data volume, intelligent processing and rapid demand iteration, and the relationship between historical measurement data and quality prediction is relatively complex, which is difficult to be represented by traditional linear models. These conditions make it difficult for general quality prediction methods to accurately reflect the quality trend of signal processing software. According to a large amount of historical data, the demand for establishing a more targeted signal processing software quality prediction model is increasingly prominent.
[0005] The passive fault event-triggered quality control methods based on personal experience are no longer sufficient to meet the increasingly scientific and intelligent development needs of signal processing software, as well as the refined management requirements of various national standards for software quality. Signal processing software quality management must adhere to the principles of prevention first, process control, and continuous improvement. The improvement of signal processing software quality prediction methods is urgently needed. Summary of the Invention
[0006] To address the aforementioned problems in the existing technology, this invention provides a signal processing software quality prediction method based on neural networks, comprising the following steps: Constructing a quality model for signal processing software: The metrics for the signal processing software quality model are determined using the GQM method; Determine the timing for predicting the quality of signal processing software, using the metrics of the signal processing software quality model at the prediction timing as the prediction index, and the quality characteristics and their subordinate quality sub-characteristics as the prediction targets. Construct a signal processing software quality prediction model based on the prediction target.
[0007] Constructing a signal processing software quality model includes: The initial hierarchical software quality model framework is used as the signal processing software model framework. Based on the signal processing software model framework, this paper integrates software quality standards, sorts out the characteristics of the signal processing software quality model under the current standards, and forms a software quality model prototype. A survey of signal processing software quality characteristics was conducted using questionnaires. Based on the survey results, the quality characteristics and sub-characteristics of the software quality model prototype were adjusted to obtain the signal processing software quality model.
[0008] The software quality metrics determined according to the GQM method include: The measurement target is determined based on the quality model of the signal processing software. The measurement target is a quality sub-characteristic under the quality characteristic. Identify key measurement questions based on the core characteristics of the measurement objectives; The metrics for the measurement objectives are determined based on the key measurement questions.
[0009] Determining the timing for signal processing software quality prediction includes: A survey was conducted on typical problems with signal processing software quality, and the requirements analysis phase was identified as the opportune time for predicting signal processing software quality.
[0010] Constructing a signal processing software quality prediction model includes: The prediction metrics are standardized to obtain a standardized dataset, which is then randomly divided into a training set, a validation set, and a test set. BP neural network models are established for the quality characteristics and quality sub-characteristics of the prediction target, respectively. Each BP neural network model has an output of 1 and an input of the number of metrics corresponding to the prediction target. The mean squared error is used as the loss function. The BP neural network models are optimized and iteratively trained by inputting the training set. The model parameters are adjusted by using the validation set until the number of iterations is reached or the model converges, thus obtaining the initial signal processing software quality prediction model. The test set is input into the initial signal processing software quality prediction model, and the signal processing software quality score generated by the model is output. The prediction accuracy of the initial signal processing software quality prediction model is determined based on the true value of the signal processing software quality. When the prediction accuracy is greater than a set threshold, the model is determined to be the final signal processing software quality prediction model.
[0011] The true quality of signal processing software is determined by scoring quality characteristics and quality sub-characteristics.
[0012] The true quality of signal processing software is determined by scoring quality characteristics and quality sub-characteristics, and also includes: The survey experts scored the data according to the criteria of excellent (5 points), good (4 points), average (3 points), poor (2 points), and indeterminate (1 point). Invalid data that could not be determined were eliminated, and the valid scores were normalized to the range of 1-5 points as the true values.
[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. A quality model and measurement system for signal processing software was constructed. By combining a hierarchical model framework with the GQM method to determine the metrics, a structured and standardized representation of software quality characteristics and sub-characteristics was achieved, providing a clear indicator basis for quality prediction.
[0014] 2. A signal processing software quality prediction method based on neural networks is proposed. By using a BP neural network to establish prediction models for quality characteristics and sub-characteristics respectively, it can effectively handle complex nonlinear relationships in quality data and improve the accuracy and reliability of prediction.
[0015] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0016] Figure 1 This is a flowchart of a signal processing software quality prediction method based on neural networks provided in an embodiment of the present invention; Figure 2 This is an initial hierarchical quality software model provided in an embodiment of the present invention; Figure 3 This is a software quality model prototype provided in an embodiment of the present invention; Figure 4 This is a signal processing software quality model provided in an embodiment of the present invention. Detailed Implementation
[0017] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the following detailed description of the solution according to the present invention is provided in conjunction with the accompanying drawings and specific embodiments.
[0018] The foregoing and other technical contents, features, and effects of the present invention will be clearly presented in the following detailed description of specific embodiments in conjunction with the accompanying drawings. Through the description of the specific embodiments, a more in-depth and concrete understanding can be gained of the technical means and effects adopted by the present invention to achieve its intended purpose. However, the accompanying drawings are for reference and illustration only and are not intended to limit the technical solutions of the present invention.
[0019] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations are intended to cover non-exclusive inclusion, such that an article or apparatus comprising a list of elements includes not only those elements but also other elements not expressly listed. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or apparatus that includes said element.
[0020] like Figure 1 As shown, an embodiment of the present invention provides a signal processing software quality prediction method based on neural networks, comprising the following steps: Step 1: Construct a quality model for signal processing software.
[0021] The initial hierarchical software quality model framework is used as the signal processing software model framework, such as... Figure 2 As shown, the initial hierarchical quality software model has a three-level structure: the first level is the overall software quality, the second level is the quality characteristics, and the third level is the sub-characteristics under the quality characteristics.
[0022] Based on the signal processing software model framework, this paper integrates software quality standards, analyzes the characteristics of signal processing software quality models under current standards, and forms a model such as... Figure 3 The software quality model prototype shown.
[0023] A survey of signal processing software quality characteristics was conducted using questionnaires. Based on the survey results, the quality characteristics and sub-characteristics of the software quality model prototype were adjusted to obtain the signal processing software quality model. Specifically: A questionnaire was developed based on the quality characteristics and sub-characteristics of the software quality model prototype. The questionnaire was scored according to the following criteria: not important (1 point), moderately important (2 points), relatively important (3 points), very important (4 points), extremely important (5 points), and uncertain (0 points). The survey experts conducted the survey and scored the quality characteristics and sub-characteristics of the software quality model prototype.
[0024] Questionnaires containing uncertainties are removed. After removing these, for the remaining k questionnaires, any quality characteristic or sub-characteristic of the j-th questionnaire is considered. In the The relative weights in the questionnaires are:
[0025] in, For this feature The score in the j-th questionnaire, where n represents the total number of quality characteristics or sub-characteristics under the quality characteristics.
[0026] The final relative weights are:
[0027] Deleting or adjusting quality characteristics with a weight less than 1 / 15 or their sub-characteristics, as shown below. Figure 4 The signal processing software quality model shown includes 6 quality characteristics and 18 quality sub-characteristics.
[0028] Step 2: Determine the metrics for the signal processing software quality model using the GQM method.
[0029] The measurement objectives are determined based on the signal processing software quality model. These objectives are sub-characteristics of the quality characteristics. Key measurement questions are determined based on the core characteristics of the measurement objectives, and the metrics for these objectives are then determined based on these key measurement questions. Specifically, the six quality characteristics are each decomposed into multiple sub-characteristics. The key measurement questions for each quality characteristic are the set of questions for its multiple sub-characteristics, and the metrics for each quality characteristic are the set of metrics for its multiple sub-characteristics. As shown in Table 1, the core characteristics of the measurement objectives define the meaning of the measurement objectives.
[0030] Table 1: Measurement Objectives and Key Measurement Issues
[0031] Step 3: Determine the timing for signal processing software quality prediction, using the metrics of the signal processing software quality model at the prediction timing as the prediction index, and the quality characteristics and their subordinate quality sub-characteristics as the prediction targets.
[0032] The software development lifecycle, based on the waterfall model, includes the requirements analysis phase, system design phase, implementation phase, integration and testing phase, and deployment and maintenance phase. A questionnaire survey was conducted to predict the timing of typical issues in signal processing software quality over the past three years. The requirements analysis phase, which had the highest percentage of typical issues identified in the survey, was determined to be the optimal time for predicting signal processing software quality.
[0033] Step 4: Construct a signal processing software quality prediction model based on the prediction target.
[0034] Specifically, the prediction metrics are standardized using Z-score to obtain a standardized dataset, which is then randomly divided into a training set, a validation set, and a test set.
[0035] Backpropagation (BP) neural network models were established for the six quality characteristics and 18 quality sub-characteristics of the prediction target, respectively. Each BP neural network model outputs 1, and the number of inputs is the number of metrics corresponding to the prediction target. The mean squared error was used as the loss function, and the BP neural network models were iteratively trained and optimized using the training set. The model parameters were then adjusted and validated using the validation set until the required number of iterations was reached or the model converged, resulting in the initial signal processing software quality prediction model.
[0036] The test set is input into the initial signal processing software quality prediction model, and the signal processing software quality score generated by the model is output. The prediction accuracy of the initial signal processing software quality prediction model is determined based on the true value of the signal processing software quality. When the prediction accuracy is greater than a set threshold, the model is determined to be the final signal processing software quality prediction model. The preferred set threshold is 0.85.
[0037] According to the empirical formula:
[0038] Where m is the number of hidden layer neurons, n is the number of inputs, l is the number of outputs, and α is a constant. When the prediction accuracy is the highest, there are 2 hidden layers and 8 hidden layer neurons.
[0039] The true value of the signal processing software quality is determined by scoring the quality characteristics and quality sub-characteristics. The survey experts score according to the scoring criteria of excellent (5 points), good (4 points), medium (3 points), poor (2 points), and indeterminate (1 point). Invalid data that cannot be determined is eliminated, and the valid scores are normalized to the range of 1-5 points as the true value.
[0040] Metric metadata from 110 software projects was collected as the training set input. Metric metadata from 4 software projects was randomly selected as the validation set, 4 software projects as the test set, and the remaining 102 software projects as the training set. After model training, the test set from the dataset was used as input to the model to obtain a set of evaluation results, which were then compared with the true values.
[0041] To verify the effectiveness of this invention, three signal processing software projects developed within the past two years, covering typical business applications, were selected: Software 1, Software 2, and Software 3. These were applied and verified using a software quality prediction model, and the results are as follows: 1) As shown in Table 2, the signal processing software quality prediction model was used to predict the six quality characteristics. The relative error between the prediction results and the true values was obtained. The average value and accuracy of the six relative errors were calculated. The maximum error was 25.56%, the lowest accuracy was 74.44%, the average relative error was 10.58%, and the average accuracy was 89.42%.
[0042] Table 2: Comparison of Predicted Results and Actual Values for Six Quality Characteristic Models
[0043] 2) As shown in Table 3, the signal processing software quality prediction model was used to predict 18 quality sub-characteristics. The relative error between the prediction results and the true values was obtained. The average value and accuracy of the 18 relative errors were calculated. The maximum error was 25.64%, the lowest accuracy was 74.36%, the average relative error was 11.11%, and the average accuracy was 88.89%.
[0044] Table 3: Comparison of Predicted Results and Actual Values of Sub-characteristic Models for 18 Quality Characteristics
[0045] This invention discloses a signal processing software quality prediction method based on neural networks. First, a hierarchical quality model for signal processing software is constructed. Quality characteristics and their sub-characteristics are adjusted and optimized by integrating software standards and questionnaire surveys. Based on the GQM method, measurement targets, key measurement issues, and corresponding metrics are determined from the quality model. Using the metrics under the requirements analysis stage as prediction indicators and the quality characteristics and sub-characteristics as prediction targets, a BP neural network model is constructed for each prediction target. Through dataset partitioning, standardization, model training, verification, and testing, when the prediction accuracy exceeds a set threshold, a final usable quality prediction model is obtained.
[0046] This invention overcomes the limitations of general quality models in signal processing software by constructing a specialized quality model and measurement system; it improves the accuracy and reliability of predictions by utilizing neural networks to process complex nonlinear relationships in quality data; it implements quality prediction during the requirements analysis phase of the software lifecycle, shifting the focus of quality control upstream, which helps prevent potential defects and reduce maintenance costs; and it combines expert experience with fixed thresholds to ensure that the prediction model is objective, reliable, and engineering-practical, providing a systematic solution for the quality management of signal processing software.
[0047] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.
Claims
1. A signal processing software quality prediction method based on neural networks, characterized in that, Includes the following steps: Constructing a quality model for signal processing software: The metrics for the signal processing software quality model are determined using the GQM method; Determine the timing for signal processing software quality prediction, use the metrics of the signal processing software quality model at the prediction timing as prediction indicators, and use quality characteristics and quality sub-characteristics under quality characteristics as prediction targets. Construct a signal processing software quality prediction model based on the prediction target.
2. The signal processing software quality prediction method based on neural networks according to claim 1, characterized in that, The constructed signal processing software quality model includes: The initial hierarchical software quality model framework is used as the signal processing software model framework. Based on the signal processing software model framework, this paper integrates software quality standards, sorts out the characteristics of the signal processing software quality model under the current standards, and forms a software quality model prototype. A survey of signal processing software quality characteristics was conducted using questionnaires. Based on the survey results, the quality characteristics and sub-characteristics of the software quality model prototype were adjusted to obtain the signal processing software quality model.
3. The signal processing software quality prediction method based on neural networks according to claim 1, characterized in that, The determination of software quality metrics based on the GQM method includes: The measurement target is determined based on the quality model of the signal processing software, and the measurement target is a quality sub-characteristic under the quality characteristic; Identify key measurement questions based on the core characteristics of the measurement objectives; The metrics for the measurement objectives are determined based on the key measurement questions.
4. The signal processing software quality prediction method based on neural networks according to claim 1, characterized in that, The timing for determining the signal processing software quality prediction includes: A survey was conducted on typical problems with signal processing software quality, and the requirements analysis phase was identified as the opportune time for predicting signal processing software quality.
5. The signal processing software quality prediction method based on neural networks according to claim 1, characterized in that, The constructed signal processing software quality prediction model includes: The prediction metrics are standardized to obtain a standardized dataset, which is then randomly divided into a training set, a validation set, and a test set. BP neural network models are established for the quality characteristics and quality sub-characteristics of the prediction target, respectively. Each BP neural network model has an output of 1 and an input of the number of metrics corresponding to the prediction target. The mean squared error is used as the loss function. The BP neural network models are optimized and iteratively trained by inputting the training set. The model parameters are adjusted by using the validation set until the number of iterations is reached or the model converges, thus obtaining the initial signal processing software quality prediction model. The test set is input into the initial signal processing software quality prediction model, and the signal processing software quality score generated by the model is output. The prediction accuracy of the initial signal processing software quality prediction model is determined based on the true value of the signal processing software quality. When the prediction accuracy is greater than a set threshold, the model is determined to be the final signal processing software quality prediction model.
6. The signal processing software quality prediction method based on neural networks according to claim 5, characterized in that, The true quality of the signal processing software is determined by scoring the quality characteristics and quality sub-characteristics.
7. The signal processing software quality prediction method based on neural networks according to claim 5, characterized in that, The set threshold is 0.
85.
8. The signal processing software quality prediction method based on neural networks according to claim 6, characterized in that, The true quality of the signal processing software is determined by scoring quality characteristics and quality sub-characteristics, and also includes: The survey experts scored the data according to the criteria of excellent (5 points), good (4 points), average (3 points), poor (2 points), and indeterminate (1 point). Invalid data that could not be determined were eliminated, and the valid scores were normalized to the range of 1-5 points as the true values.