Hyperspectral Image Classification

The objective of the neural network is to transform the input to meaningful output. Neural networks are often used for statistical analysis and data modeling. Neural network has many uses in data processing, robotics, and medical diagnosis [2]. From the starting of the neural network there are various types found, but each and every types has some advantages and disadvantages. Deep learning and -neural network software are the categories of artificial neural network. The parallel process also allows ANNs to process the large amount of data very efficiently. The artificial neural network is built with a systematic

These 16 features included 12 features calculated based on the 6 multispectral bands, which is mean value and standard deviation of these bands. In addition, we chose intensity, texture-variance, texture-mean, and NDVI (Normalized Difference Vegetation Index) for classifications. Finally, training samples were selected for each classification category based on the previously segmented and merged objects

With the use of PCA, it became difficult to determine the spectral signature of objects for a long time. In the new data cube a “pixel profile” is not a spectral signature.

Thus our proposed optimal feature subset selection based on multi-level feature subset selection produced better results based on number of subset feature produced and classifier performance. The future scope of the work is to use these features to annotate the image regions, so that the image retrieval system can retrieve relevant images based on image semantics.

The training data contained both labeled data D_la={〖x_i,y_i}〗_(i=1)^kl and unlabeled data D_un= {〖x_j}〗_(j=kl+1)^(kl+u) where x_(i ) is the feature descriptor of image I and y_i={1,…,k} is its label .k is the number of categories. l is the number of labeled data in each category, and u is the number of unlabeled data. Our method aims to learn a high-level image representation S by exploiting the few labeled data D_land great quantities of unlabeled ones, which is then fed into different classifiers to obtain final classification results. The procedure of semisupervised feature learning by SSEP is shown in Fig. 1. First, a new sampling algorithm based on GNA [19] is proposed to produce T WT sets P^t={(〖s_i^t,c_i^t)}〗_(i=1)^kp , t ∈{1,…..,T}

We have used support vector machine (SVM) for classification task. We have used RBF kernel for training the classifier. 10 fold cross-validation is used for determining cost parameter C and best kernel width for RBF kernel function. If we perform classification without any feature selection or feature extraction then the accuracy is 48.99% and 65.82% for AVIRIS and HYDICE image respectively which is very poor and it highly motivates us to apply feature reduction technique. In table II we have shown the classification accuracy for each of the pair of class for PCA, MI and PCA-QMI.

The particular finding parameters (i.e. threshold, fwhm, sharpness, and roundness) were tuned to the data to maximize true identification yields as determined by visual inspection. Some parameters like fwhm and signal threshold were set adaptively by using image based diagnostics like a first order estimate of the Gaussian fit PSF and the background noise levels, respectively. It was particularly tricky to find parameters that worked well for both clear and highly nebulous regions with small features. Also, the parameters had to be tuned for particular filters and co-adding depths. The subsequent photometry routines for grouping, nstar fitting, and PSF subtraction were all part of the Photoutils package. A sigma clipped median was used to estimate the

The first way is fusing the standard camera data stream with that of a forward looking infrared (FLIR) system. FLIR systems have been used for years by the military to detect enemy combatants and vehicles via their heat signatures. The major advantage of this approach is that it works just as well, if not better, during the night. While a standard camera produces a large amount of low-light noise, infrared cameras can produce high resolution images which can be used for accurate obstacle detection. I plan to use the fused data for feature extraction with a suite of image processing methods, such as Canny edge detection. Finally, I plan to use the extracted features to classify them with common methods such as neural networks or SVMs. I expect this phase of my research to take one to one and a half years, because it relies on proven technology being used in a new

I have started working on a research project using machine learning to build an intrusion detection mechanism. To build up on this background, I am also interested in the projects of implementing machine learning algorithms with an application toward the biological sciences and machine learning from multi-modal data. Working on designing machine learning algorithms with Dr. Kevin Chen would be an enlightening experience to build up my ability with the burgeoning field of machine learning. Machine learning from multi-modal data would also be useful as physical security is also a crucial component of cyber-security as well. I hope to contribute to designing the data sets and machine learning algorithms that can handle the new sensory data available to computers.

The purpose of this paper is to provide a unique and concise solution to a biomedical problem from a supervised learning and statistical perspective. As the solution is purely from a statistical perspective without support from proper biomedical theories, the solution might not serve practical or meaningful biomedical values. In our solution, we have reduced feature dimension to as low as 0.1%. The selected features served as biomarkers might be useful to the diagnosis of LGL leukemia, but it requires further analysis and proof in the field of biomedical

However, all have their advantages and pitfalls. Spectral matching based methods provide high detection accuracy but not suitable for camouflage detection due to unavailability of spectral signatures of camouflaged targets. Spectral anomaly and ICA based methods have been reported mostly on synthetic and Visible-SWIR Hyperspectral data with varying performance accuracy depending upon the level of spectral variability13, 21. The real-world conditions always differ from synthetic data due to nonlinear atmospheric attenuation, background clutter and sensor noise. Moreover, there exists the poor spectral contrast between the camouflaged objects and natural background. As a result, target detection accuracy drops down and false alarm rate increases. Therefore, there is a need to develop a method for efficient and robust detection of camouflaged targets in MWIR spectral region, which is mostly used in reconnaissance and surveillance

Remote sensing is capturing images of an object at some distance away in a bird-eye view with a sensor.  With the increasing popularity of satellites, automated data analysis is introduced. However, human beings must still be engaged in the analysis process because they have the ability to understand the data, create concepts from it, make decisions and solve problems. With that being said, the analysis process requires experience, which improves perceptual learning. With experience and practice, analysts are able to derive information immediately and more effectively, incorporate high-level concepts and identify important features that a non-expert might miss. In the process of interpreting satellite or radar imagery, human beings rely on

IV. PATTERN RECOGNITION USING NEURAL NETS For really complex problems neural networks outperform their competition. With the aid of GPU’s [1], the neural networks can be trained faster than ever before. Deep learning is specially used to train computers to recognize patterns.

To construct an optimal hyperplane, SVM employs an iterative training algorithm, which is used to minimize an error function. According to the form of the error function, SVM models can be classified into four distinct groups:

Principal component analysis (PCA) is a statistical method for correlation, from a small number of independent principal components to represent the total samples. It is always applied in geochemical exploration for extraction anomalies from element variates and integration of multi-source geo-information. In a recently year, several improved PCA methods were proposed by traditional PCA. Cheng (2006) proposed spatially weighted principle component analysis (SWPCA) for images processing by spatially weighted correlation coefficient matrix by incorporating a fuzzy mask defined by sample location.