Testing Deep Learning-based Visual Perception for Automated Driving

Due to the impressive performance of deep neural networks (DNNs) for visual perception, there is an increased demand for their use in automated systems. However, to use deep neural networks in practice, novel approaches are needed, e.g., for testing. In this work, we focus on the question of how to test deep learning-based visual perception functions for automated driving. Classical approaches for testing are not sufficient: A purely statistical approach based on a dataset split is not enough, as testing needs to address various purposes and not only average case performance. Additionally, a complete specification is elusive due to the complexity of the perception task in the open context of automated driving. In this article, we review and discuss existing work on testing DNNs for visual perception with a special focus on automated driving for test input and test oracle generation as well as test adequacy. We conclude that testing of DNNs in this domain requires several diverse test sets. We show how such tests sets can be constructed based on the presented approaches addressing different purposes based on the presented methods and identify open research questions.

Mathematical Planning When Choosing Rational Dosages of Ingredients for Adjusting the Composition of Bakery Products

The article provides a mathematical solution to the problem of choosing the dosages of ingredients in regulating the baking properties of wheat flour with the use of soy protein isolate, which allows not only adjusting the process but also improving the biological value of bakery products. The chosen main factors that influenced bread quality were the following: x1 – the dosage of soybean isolate, %; and x2 – dough moisture, %. These factors were compatible and uncorrelated. The central-composite rotatable design was chosen. The research program was laid out in the experiments planning matrix. The order of experiments was randomized using the table of random numbers, which excluded the influence of uncontrollable parameters on the results of the experiment. For statistical processing of the experiment results, the following statistical criteria were used: the significance of regression equations coefficients – Student's t-test, adequacy of equations – Fisher's variance ratio. As a result of the experimental data statistical processing according to the known methods, regression equations that adequately describe at the significance level of 5 % the dependence of the specific volume of bread and dimensional stability on the studied factors were obtained.

The Layout of Maize Variety Test Sites Based on the Spatiotemporal Classification of the Planting Environment

An appropriate layout of crop multi-environment trial (MET) sites is imperative for evaluating new crop varieties’ performance in terms of agronomic traits and stress tolerance, and this information is used to determine the utilization value and suitable promotion region of new varieties. Actually, traditional maize test sites have been selected according to the experience of breeding experts, which leads to the strong subjective and unscientific conclusions regarding sites, as well as test results that are not representative of the target population of environments (TPE). Therefore, in this study, we proposed a new method for MET sites layout. Meteorological data, maize growth period data, and county-level maize planting area data were collected for the spatiotemporal classification of a given maize planting region to analyze change rules in the environmental category of each minimum research unit within the study period. If the occurrence frequency of its final attribution category reaches a certain threshold (50%), this minimum research unit is classified as a typical environment region; otherwise, it is classified as an atypical environment region. Then, the number of test sites in each environmental category is allocated by spatial stratified sampling. At last, we establish the optimal test sites layout and a reliability measurement (test adequacy) methods. The practicability of this method was proved by taking the Three Northeastern Provinces of China as the study area. The result shows that there should be 112 test sites in the study area, the distribution of the test sites is uniform, and the environmental representation is high. Test adequacy analysis of the test sites reveals that most of the environmental categories have a test adequacy that reaches 1 in each test period. The method proposed in this paper provides support for the scientific layout of crop varieties test sites and helps to improve the representative and reliability of variety test results while optimizing resources.

A Multi-Agent-Based Approach for Critical Components Identification and Testing

In real time software systems, testing plays a crucial role as any of the critical components in these systems are left undetected, then inadvertent effects will happen which will lead to erroneous operations, system failure, high cost and resource wastage etc. To address this most important and the emergent problem, this research work proposes an effective method by means of multi-agents based approach to identify such critical components and execute test cases along the critical test paths which will aid in effectively covering them during testing. Finally, this paper also compared the performance with existing approaches in terms of time taken for the search process and the component coverage based test adequacy criterion to ensure quality of the software.

Research on Testing Adequacy Criterion of Reactor Physics Code of NESTOR

In view of the characteristics of the physical code Nestor the focus is on the correctness of calculation for which the test adequacy criterion has been established. This is based on structural coverage and the input domain. According to such test adequacy criterion, testing strategies have been applied on the entire testing process. They consist of unit static, unit dynamic, integration, system and regression test strategy. Each strategy is composed of test target, test range, technology and method, entry criterion, completion criterion, test focus and priority. After compared with 11 basic benchmarks from nuclear power plants and calculation result of benchmark programs, the ELEMENT program result is correct and credible; the relative error of result is less than three percent. The ELEMENT testing is adequacy. Its test cases covers fuel grid element types, fuel types, non-combustible grid element types, and control rod computational models. Furthermore, it puts forward a research direction in the future.