Mesopic Visual Performance of Cockpit’s Interior based on Artificial Neural Network

The ambient light of cockpit is usually under mesopic vision, and it’s mainly related to the cockpit’s interior. In this paper, a SB model is come up to simplify the relationship between the mesopic luminous efficiency and the different photometric and colorimetric variables in the cockpit. Self-Organizing Map (SOM) network is demonstrated classifying and selecting samples. A Back-Propagation (BP) network can automatically learn the relationship between material characteristics and mesopic luminous efficiency. Comparing with the MOVE model, SB model can quickly calculate the mesopic luminous efficiency with certain accuracy.


A. Cockpit's Interior Ergonomics
Visual comfort occupies an increasing important place in our everyday life, but also in the field of aeronautics which is the subject of this paper.Modern science and technology is people-oriented.More and more human factors were taken into consideration on the development of modern civil airplane.
A subject of applying ergonomics into man-machine relationship develops gradually, which has brought about more and more attention.The ambient light of cockpit consists of natural sunlight, instruments panels, inside lighting systems and interior's reflecting light.The quality of ambient light in the man-machine system should meet the requirements, as well as providing people visual information about activity both in quality and quantity.It should meet ergonomics requirements of perceptive information, aiming at making people comfortable and pleasant.The cockpit's comfort plays an important role in pilot's job, especially the visual performance.About 80% outside information is received though vision, which makes vision the most important channel to communicate with external world [1].Visual comfort is the psychological feeling about comfort level in ambient light.So, visual comfort is a quantity of psychological feeling.
Uncomfortable vision will cause a series of symptoms, usually appears as redness and swelling, pain, itching, tears, dizziness or even intestines and stomach problems.Comfortable environment of cockpit would guarantee the pilot keep a normal state in the process of work, to avoid flight accidents caused by visual factors.
In the man-machine ergonomics study, establish the inner relationship between different materials properties and light source, and the received luminance, light intensity, contrast and color in the specific conditions.Simulate different characteristics of efficiency of different light source and different materials.The influences of these multidisciplinary factors are not independent.They're of complicated nonlinear relation.Past studies domestic and oversea are mostly aimed at single factor of variables, without considering the nonlinear relationship between multiple factors and the coupling mechanism.
The established design standards and norms cannot completely meet the pilot's ergonomics requirements in the real flight environment, thus increasing the design difficulty of a cockpit ergonomics system.There's a long history according to the visual ergonomics research in cockpit.Britain and America have already done a lot of experimental and theoretical research, and established design standards of illumination and colorimetry.However, there's only a little study for the interior system of ergonomics problem.The domestic related research is scattered, mainly paused in the qualitative subjective evaluation level, which cannot form a systematic theory.Through the research of aircraft cockpit's coupling mechanism, characteristics of the pilot's visual perception and comprehensive effect with multivariable factors, we can build the civil aircraft cockpit's interior ergonomics theory model and application mechanism.OPTIS established a cockpit's visual system model [2] by CATIA based on ergonomic design criteria and a certain type of aircraft cockpit, which is shown in Figure .1.Then various optical properties could be set, including the characteristics of www.ijacsa.thesai.orglight source spectral and optical materials which participating in the process of light transmission.After that, they can be applied to the optical tracking system to simulate the light process.
For safety reasons, visual information must be seen as comfortable as possible by the aircraft pilot in any light conditions.In this paper, we focus on the cockpit's mesopic visual performance based on the ANN method.

B. Mesopic Vision
Mesopic light levels are those between the photopic (daytime) and scotopic (extremely low) light levels.The mesopic luminance range covers brightness between about 0.001 and 3 cd/㎡.Most night-time outdoor and traffic lighting environments and some indoor lighting are in the mesopic range [3].
As human eyes have different perceptions on light fusions from different frequency, there comes to be different brightness, for observers, between lights of different wavelength even with the same power.Luminous efficiency function curves indicate such a human eye character.

Photopic luminous efficiency function
which is fit for delineating the spectral response within a 2degree range of human eyes in a higher brightness, is the most widely used function in this field and was brought forward on the 6th conference of CIE in 1924.CIE has successively brought forward the function with a wider range of 10 degree and also the luminous efficiency function  A number of studies of visual performance at mesopic light levels have been conducted, which underscore the importance of recognizing the distinction between photometry and a complete characterization of visual responses at mesopic levels [4].The EC project MOVE [5] (Mesopic Optimisation of Visual Efficiency) was carried out during 2002-2004 in the EC Fifth Framework programme (G6RD-CT-2001-00598).The objective of the project was to define relevant spectral sensitivity functions for the luminance range of 0.01 -10 cd/ ㎡, where standardisation is most urgently needed.The TC1-58 Technology Committee found in 2000 also by CIE came up with a better model using the method of visual performance and such a function below was brought forward [6]: Where   m V  represents the mesopic luminous efficiency function under the environment of a certain backdrop brightness, parameter which is located between 0 and 1 based on backdrop brightness and spectral power.x=1 is in photopic conditions while x=0 in scotopic conditions.
Illumination is defined as the transparent flux on unit area, and flux is available from function below:  In this paper, a new method has been used into human vision model to simplify the nonlinear relationship between light characteristics and mesopic vision.In real time scenario, there are many factors taken into consideration.However, we focus on some dominant variable in this paper.

II. ARTIFICIAL NEURAL NETWORK
An artificial neural network (ANN), usually called neural network (NN), is a mathematical model or computational model that is inspired by the structure and/or functional aspects of biological neural networks.A neural network consists of an interconnected group of artificial neurons, and it processes information using a connectionist approach to computation.
From 1940s, with human being's fully understanding of the brain structure, composition and the most basic unit, Artificial Neural Network was arose [7].Simplified model www.ijacsa.thesai.org(ANN) was established after combining mathematics, physics and information processing method, and making neural network abstracted.
As an active marginal subject, the research and application of neural network is becoming a hot pot of artificial intelligence, cognitive science, neurophysiology, nonlinear dynamics, and other related subjects.In last ten years, academic research according to neural network is very active, and puts forward almost a hundred of neural network model.Neural network is also widely used in analysis of input and output with multiple variables.In the process of aircraft design, using the neural network to optimize pneumatic parameters has obtained some progress.
In this paper, SOM network is used to compress a set of high-dimensional input parameters that contain material characteristics onto a two-dimensional SOM grid.SOM network is different from other artificial neural networks in the sense that it uses a neighborhood function to preserve the topological properties of the input space.The neurons will classify the space, each neuron representing a partition of that space.
SOM network employs an Winner-Takes-All (WTA) operation, which only the winner is allowed to adjust the weight connecting to the input.The training process includes sequential steps [8].Nt , and all neuron weights in the radius will be adjusted by (2).
Where the Learning Rate  will be the function of training time t and the radius of neighbor N    

III. EXPERIMENTAL RESULTS
To study the mesopic vision performance of pilot in cockpit, we gathered measurement variable into an initial database, shown as in Figure 5.In the condition of experiment, input variable including material reflectivity, material transmissivity, material absorptivity, luminous flux, color temperature, material coordinate x and material coordinate y, as well as output variable including brightness, light intensity, contrast, color x and color y.

A. Dominant Wavelength
Calculate the dominant wavelength with x-y chromaticity in a Chromaticity Diagram [9], just illustrated in Figure 7.
Construct a line between the chromaticity coordinates of the white point on the diagram (for instance, CIE-E) and the chromaticity coordinates, and then extrapolates the line from the end that terminates at the filter point.The wavelength associated with the point on the horseshoe-shaped curve at which the extrapolated line intersects is the dominant wavelength.Table 1 gives some common illuminants used as a white reference.

B. Mesopic Luminous Efficiency
Then the photopic luminous efficiency and the scotopic luminous efficiency could be obtained by dominant wavelength.The parameter x in formula (1) could be calculated according to Table 4 in reference [5].
According to the brightness and s/p ratios, x could be approximately calculated in certain conditions as shown in Table 2.The s/p ratio is chosen in the condition of typical overcast sky, which is 2.36 [10].
Linear fitting function's used to fulfill the x-value.The MOVE model is applicable for other conditions such as typical sunlight sky and typical direct sunlight sky.

C. SOM Network Results
According to the design requirement, a total of 12 samples are selected, including brightness, color x and color y.   4 shows the resulting SOM with cluster groups considering the all 3 characteristics.Figure 9 demonstrates the responses to different samples.The number is ranked from left to right and up to down, sign as 001 to 012.The more similar the characteristics of samples are, the closer the colored maps are.
From the colored result, we can easily get that most of the samples performed like the same in group 1. No.012, 014 and 015 are the most distinguishing ones.www.ijacsa.thesai.org

D. BP Network Results
Construct a BP network, with color coordinate and brightness as input and mesopic luminous efficiency as teacher signal (target).The number of hidden nodes depends on the number, scale and complexity of samples.To confirm the number of hidden nodes, we take the formula as follow: where m is the number of hidden nodes, n is the number of input nodes, l is the number of output nodes, and  is a constant between 1 and 10.
The main parameters of BP network are of

CONCLUSIONS
This paper put forward a SB model, which uses SOM network to classified different samples into different groups.Chose a certain group, the training result is better than before.Use BP network to simplify the relationship between the mesopic luminous efficiency, and the different photometric and colorimetric variables in the cockpit.After comparing with the MOVE model, SB model takes advantage of ANN to simplify the relationship, and convenient to calculate mesopic luminous function, as well as has more concentrate error density distribution.To make research more accurate, we will take more research on simulating human's eye, and construct corresponding model.Photopic vision and scotopic vision are taken into consideration in the next as well.

Figure 1 .
Figure 1.Cockpit's visual system model environments of low brightness less than 0.001cd/㎡.

Figure 2 .
Figure 2. Photopic luminous efficiency function   V  and scotopic 2) Where x  stands for the total flux,   P  is the spectral arrangement function of light source,   x V  is the luminous efficiency function on certain brightness, x M is the normalization factor.// sp sp   .

Figure 3 .
Figure 3.The process of calculating responses by (4) to all the neurons on competitive layer.Find the winning neuron * : Calculate the radius of neighbor neurons   * j d) Where, k t is the teacher signal and k o is the output of neuron k .e) For each neuron on hidden layer ( 21 lL  the network weights as follows: to step b) for another input sample until the termination condition is met.

Figure 5
Figure 5.Initial database After observing, material reflectivity, material transmissivity, and light intensity are invalid.In the condition of color temperature equals 5000K and luminous flux equals 1000lm.Initial database can be trimmed.The sample database shows as Figure 6.

Figure 7 .
Figure 7. CIE 1931 Chromaticity Diagram Chromaticity coordinates of some common illuminants used as a white reference.
value of 1, shown as in Figure8.

Figure 8
Figure 8. x-value and

For
test result, total cycles of training m t is defined as 20000.

Figure 9 .
Figure 9. 2d-response by SOM After the whole training, all 12 samples are classified into 4 groups on a 22  map.Table 4 shows the resulting SOM

Figure 10 .
Figure 10.Error density distribution function After the whole training, the result is shown in Figure 10.The black square line represents error density distribution function of MOVE model, as well as the red circle line represents error density distribution function of SB model.Observing the peak of these two models, mesopic luminous efficiency function of SB model is more concentrated than MOVE model.

TABLE I .
After calculation, the dominant wavelengths of color coordinates are shown in

TABLE II .
DOMINANT WAVELENGTH OF COLOR COORDINATES