Neural Networks for regression Analysis

The Study of Quantitative Estimation Models of Leaf Area Index Based on High Resolution Remote Sensing Image
Full-size image (452 K)

Forest resources are declining with the population growth, therefore, to access to the information about regional and global forest change accurately and in real-time is particularly important. Leaf area index(LAI) as an important indicator to understand the dynamics and balance of the plant canopy gap, can provide structured quantitative information to a description about energy exchange between the surface’s matter and energy of the plant canopy. In addition, LAI and biomass have a close relationship, and also closely related to the multi-scale ecosystem’s prolificacy. However, using the traditional methods of measurement is almost impossible to obtain the LAI in a large regional scale by a variety of influencing factors, such as definition, sampling methods, data analysis, instrumental errors and so on. Now, with the growing depth and maturity of remote sensing technology applications, especially the emergence of high-resolution remote sensing image, so that it is possible to estimate and dynamic monitor LAI accurately, non-destructively and real-timely in a large area.The vegetation of Huang Fengqiao Forest Farm in You County, Hunan Province as the object of study, in this paper to set up108plots in the study area with a typical sampling method, and to use the LAI-2000plant canopy analyzer to measure LAI values of these vegetation. Using forestry remote sensing technology and GPS measurement techniques to extract16remote sensing factors and3geographical factors of plots from WorldView-2images and1:10000topographic maps of the study area, and analyze the correlation between LAI values and factors, and then select the factors which there are significant or highly significant correlation with LAI values which as input variables of the assessment model. Using the selected factors which as independent variables to construct a statistical regression model, and then using principal component method to reduce the dimension of the selected factors which as the original data, use stepwise regression analysis method to construct multivariate statistical regression model. Using the selected factors as the network input variables and training to build BP neural network model to estimate the LAI. Using the test sample data to validate and evaluate the models’accuracy. The results show that:(1)、 Found by correlation analysis between the19remote sensing factors and geographical factors and LAI:In addition to Blue band reflectance values (Bandl), Green band reflectance values (Band2), Slope and Aspect, there are significant or highly significant correlation between the LAI values and the other15factors, such as Red band reflectance (Band3), Near-infrared reflectance (Band4), Ratio vegetation index (RVI), Normalized difference vegetation index (NDVI), Difference vegetation index (DVI), Perpendicular vegetation index (PVI), Soil adjusted vegetation index (SAVI, L=0.10, 0.25, 0.35, 0.50), Modified soil adjusted vegetation index (MSAVI), Atmospherically resistant vegetation index (ARVI, γ=0.50, 1.00), Modified chlorophyll adjusted vege tation index (MCARI), Elevation. The correlation is negatively between Band3and LAI, correlation coefficient is-0.4231.(2)、 Found by different types of statistical regression models, which were constructed between LAI and the factors:In all linear regression models, the exponential regression model with NDVI as the independent variable (y=0.1195exp(4.9714x)) has highest estimation accuracy, is89.30%.(3)、 Found by the multiple linear regression model which using multivariate stepwise regression:The cumulative contribution rate of the first seven principal components is99.99%after principal component analysis, achieve dimensionality reduction, at the same time, only loss little amount of information of the original data. The multivariate stepwise regression model by them as the alternative variables is y=-0.9942+20.4543xi+0.3914×2, (xi is the seventh principal component (F7), X2is the fourth principal component (F4)), its estimation accuracy is87.72%, little less than the linear regression model’s.(4)、 Found by BP neural network model established by the neural network thought:The BP neural network models’estimate accuracy with different number of hidden layer’s neurons are all more than80.00%, the highest accuracy is85.83%when the number of neurons in the hidden layer is set to11.(5)、 It is feasible to quantitative estimate the LAI values in a wide range fast and accurately. The models’accuracy, have been established in the paper, are all more than80.00%, the exponential regression model has highest accuracy with NDVI as the estimator.

Source: Agricultural Science Paper

You might also like:

ATI Radar Signal Analysis and Processing using MATLAB Short Course Technical Training Sampler Video
ATI Radar Signal Analysis and Processing using MATLAB Short Course Technical Training Sampler Video
Radar and Its Applications 1962 United States Army Training Film
Radar and Its Applications 1962 United States Army Training Film
Simulation Platform Signal Processing Analysis UWB Radar Multi Fading Channels - Hebrew
Simulation Platform Signal Processing Analysis UWB Radar Multi Fading Channels - Hebrew

Monday prosper reading material...

2008-08-25 09:22:48 by grativo

1. This loan will provide the necessary capital to purchase software, marketing materials and entrepreneurial educational tool taught by master entrepreneur and CEO of Nouveau Riche University, Jim Piccolo. With the ability to now focus full-time as opposed to free-time I expect to increase my income to a respectable $1,800/month.
2. Seriously folks, this loan will be used to take my Forex Trading Strategy, which is working very well, to the next level. Look at the picture in the listing. Over 150% profit in 13 days of active, 24 hour, automated trading using my software. (Turns out never traded for real money)
3. This loan will be used to help me invest 10k into a forex managed account investment fund. I am a good candidate for this loan because I am investing my own money...

No Loss Trading Plan that really works for your.

2008-01-18 20:41:56 by bfnnrgn

Are you in the Dallas, TX area? I have a trading plan that is non speculative, non directional, and generates a very high rate of return. It works everyday - all the time - with no losses - (yes) it "really" works. I have been trading the financial markets for 30 years and I understand exactly what I am doing. As a demo, I started trading a friends Forex account of just $6,000.00 on 12/20/07 and I generated a 100% return in just three weeks using leverage with no losses. Luck was not part of the success. A properly funded brokerage account will never have losses. I know it sounds unlikely but I can demonstrate and prove it "is" true. It took me most of my life to develop this technique and it did not come cheap to me. I call it my zero sum no loss trading plan and it will beat any...

Pay one way or another

2005-09-04 14:25:42 by o-lOl-o

$2000 for an online course huh. If it works for you. A few weeks is really good to actually begin to profit in live trading. Staying consistant over the long hual is the trick now. That takes discipline. Avoid the blowout sessions where a trader chases good money after bad trying to recoup losses. Or greedily jumps into a trade with little research.
I home schooled myself over about a year period. I didn't have the money to trade at the time so it really didn't matter. I was content to learn and save till I had it. I had some expenses that was a priority over forex. When I did have the extra cash not swallowed by life I had been profiting for over 6 months on paper and the transition to live was relatively seemless.
I'm getting nice returns now for almost a year live. From...

U.S. Recession Risk Still Low In January  — Seeking Alpha
The resulting diffusion index and changes in the diffusion index are used to estimate the probit, logit and neural network forecasting models. The graph of the diffusion index from 1/1/2003 to 2/1/2013 is presented in Figure 1 below (in red - left axis).

Furuno DRS12A RADAR Sensor for NavNet 3D Networks. 12 KW, 72 NM Range
Furuno DRS12A RADAR Sensor for NavNet 3D Networks. 12 KW, 72 NM Range
Furuno DRS2D RADAR Sensor with Radome
Furuno DRS2D RADAR Sensor with Radome
Radar Pulsed Signal Analysis
Radar Pulsed Signal Analysis

Related posts:

  1. Neural Networks and regression
  2. Neural Networks regression Analysis
  3. Neural Networks for Process Control
  4. Neural networks for Stock analysis