Visualization for Explanation of Deep Learning-Based Defect Detection Model Using Class Activation Map

Recently,convolutional neural network(CNN)-based visual inspec-tion has been developed to detect defects on building surfaces automatically.The CNN model demonstrates remarkable accuracy in image data analysis;however,the predicted results have uncertainty in providing accurate informa-tion to users because of the“black box”problem in the deep learning model.Therefore,this study proposes a visual explanation method to overcome the uncertainty limitation of CNN-based defect identification.The visual repre-sentative gradient-weights class activation mapping(Grad-CAM)method is adopted to provide visually explainable information.A visualizing evaluation index is proposed to quantitatively analyze visual representations;this index reflects a rough estimate of the concordance rate between the visualized heat map and intended defects.In addition,an ablation study,adopting three-branch combinations with the VGG16,is implemented to identify perfor-mance variations by visualizing predicted results.Experiments reveal that the proposed model,combined with hybrid pooling,batch normalization,and multi-attention modules,achieves the best performance with an accuracy of 97.77%,corresponding to an improvement of 2.49%compared with the baseline model.Consequently,this study demonstrates that reliable results from an automatic defect classification model can be provided to an inspector through the visual representation of the predicted results using CNN models.

Visual Quality Analysis of Urban Park Landscapes Based on Eye Tracking: A Case Study of Nanjing Xiaohong Stone Carving Park

The landscape quality of urban parks is an important aspect of tourists'landscape experience.A full understanding of their relationship is beneficial to the park design,construction and management.In this paper,Xiaohong Stone Carving Park in Nanjing was selected as a case study.Firstly,the subjective landscape visual quality evaluation of tourists was obtained through the scenic beauty evaluation,and the objective landscape visual preference of users was obtained through eye movement experiment and analysis.Secondly,the values of various park landscape elements were measured respectively.And then the correlation between the subjective and objective evaluation and the values of park landscape elements was analyzed.The results showed that:(1)The visual preference of plant cluster landscape with beautiful shape,distinct layers and significant color contrast is high;(2)The preference for waterfront scenes with significant terrain differences,rich background vegetation levels,and clear and vivid water side reflections is high;(3)The designed artificial structures often become the focus of attention in the whole scene,which can improve the beauty of the scene;(4)The overall landscape composition is very important for the beauty of the scene.Finally,through the comprehensive analysis with the design expectation,the optimization strategy was proposed.

Quantitative Indexes of Streetscape Visual Evaluation and Validity Analysis

To establish valid indexes for streetscape visual evaluation, this paper proposed 6 quantitative indexes to refl ect objective visual properties of streetscapes based on the image processing concept, and took real digital images of 10 pedestrian streets to verify the validity. Mathematical analysis of subjective and objective indexes showed that street width/height ratio, street width, sky openness index, visual entropy and color richness index could refl ect objective visual properties of streetscape(correlation coeffi cient r > 0.633, signifi cance P < 0.05); visual complexity accounted for the highest percentage in the evaluation conclusion(22.6%), so it was the most signifi cant infl uence factor.

Visual Evaluation of Matching Degree between Fisherman Hat and Female Face

By the means of visual evaluation,the matching degree of nine types of fisherman hats and seven face shapes was studied.The characteristics of hats and faces were parameterized,and the images in which seven kinds of faces wore different fisherman hats were processed by Photoshop gray levels.Through a random questionnaire,descriptive analysis,and mean and dispersion analysis,the recommendation table of fisherman hats for different female faces was obtained,and the highly preferred and matching degree types of fisherman hats were obtained.It not only provides reference for virtual hat tests,but also gives references for parameterized hat designs from the collocation aesthetic viewpoint.

Visual Walkability in Wangjing Area from the Perspective of Young Men

This paper mainly evaluated the walkability of the streets around the subway Futong Station in Wangjing area,and with the help of relevant theoretical research conclusions such as environmental behavioral psychology,to study the effect of the visual walkability of streets in different communities on the walking activities of young male residents.Through in-depth analysis of the characteristics of streets that are suitable for walking and young male residents like to walk,practical and feasible reference suggestions were provided for the design of street public space and the establishment of a more livable urban environment.The text was divided into 3 parts:the determination of the functional types of the streets around Futong Station;the characteristics of young men's crowd activities and the establishment of an evaluation system;the visual walkability evaluation of the streets around Futong Station.

Research on the Perception of Green Visual Exposure Space Based on SD Method:Taking Saishang Old Street Cultural Area in Hohhot as an Example

With the rapid development and improvement of urban construction,the influence of green environmental exposure on mental health has attracted increasing attention,but the relationship between the psychological perception and the object index has been less explored.From the perspective of the individual and environmental perception,space optimization of cultural blocks was discussed to provide a reference for the development of other old streets.

Motivation effect of animated pedagogical agent's personality and feedback strategy types on learning in virtual training environment

Background The personality and feedback of an animated pedagogical agent(APA)are vital social-emotional features that render the agent perceptually believable.Their effects on learning during virtual training need to be examined.Methods In this paper,an explanation model is proposed to clarify the underlying mechanism of how these two features affect learners.Two studies were conducted to investigate the model.In Study 1,the effect of the APA's personality type and feedback strategy on flow experience and performance was reexamined,revealing significant effects of the feedback strategy on flow and performance and a marginally significant effect of the personality type on performance.To explore the mechanism behind these effects,a theoretical model is proposed by distinguishing between intrinsic and extrinsic motivation effects.In Study 2,the model was evaluated,and the APA's personality type was found to significantly influence the factors in the path of the extrinsic motivation effect rather than those in the path of the intrinsic motivation effect.Results In contrast,the feedback strategy affected factors in the path of the intrinsic motivation effect.Conclusions These results validated the proposed model.Further distinguishing the two motivation effects is necessary to understand the respective effects of an APA's personality and feedback features on learning experiences and outcomes.

Objective visual performance evaluation with visual evoked potential measurements based on an adaptive optics system

An objective visual performance evaluation with visual evoked potential （VEP） measurements was first inte- grated into an adaptive optics （AO） system. The optical and neural limits to vision can be bypassed through this system. Visual performance can be measured electrophysiologically with VEP, which reflects the objective func- tion from the retina to the primary visual cortex. The VEP ts without and with AO correction were preliminarily carried out using this system, demonstrating the great potential of this system in the objective visual performance evaluation. The new system will provide the necessary technique and equipment support for the further study of human visual function.

Versus—A tool for evaluating visualizations and image quality using a 2AFC methodology

Novel visualization methods and strategies are necessary to cope with the deluge of datasets present in any scientific field to make discoveries and find answers to previously unanswered questions.These methods and strategies should not only present scientific findings as images in a concise way but also need to be effective and expressive,which often remain untested.Here,we present Versus,a tool to enable easy image quality assessment and image ranking,utilizing a two-alternative forced choice methodology(2AFC)and an efficient ranking algorithm based on a binary search.The tool provides a systematic way of setting up evaluation experiments via the web without the necessity to install any additional software or require any programming skills.Furthermore,Versus can easily interface with crowdsourcing platforms,such as Amazon’s Mechanical Turk,or can be used as a stand-alone system to carry out evaluations with experts.We demonstrate the use of Versus by means of an image evaluation study,aiming to determine if hue,saturation,brightness,and texture are good indicators of uncertainty in three-dimensional protein structures.Drawing from the power of crowdsourcing,we argue that there is demand and also great potential for this tool to become a standard for simple and fast image evaluations,with the aim to test the effectiveness and expressiveness of scientific visualizations.

New trends in China's science and education evaluation research under the background of“Breaking the Five- only”

Based on the background of the special clean-up action of"Breaking the Five-only",this paper combs the relevant policies of domestic science and education evaluation.Using CiteSpace and VOSviewer scientific measurement software,this paper makes a visual analysis on the related domestic research of the"Five-only"and"science and education evaluation",and expounds the frontier hot spots and trends of science and education evaluation research in China.Based on this,this paper summarizes the countermeasures and suggestions on how to"break"the"Five-only"and how to"establish"the"new system of science and education evaluation",in order to provide a reference for the sustainable and healthy development of science and education evaluation in China.