Here is how you have to narrow down your research problem in hyperspectral remote sensing

Research on Initial Stage Common Fallacies 

 Usually, the below two common mistakes are highly possible at the initial stage of our research 

• Hopping our interests from one keyword to another
• Overnight identification of research problem

• Hopping our interests from one keyword to another

"What could be more interesting than this?"

         During my initial research journey, when I search with keyword data mining I found more interesting topics such as Business Intelligence, Image Mining, Text Mining, Social Networking Data Analysis, Video Analytics, etc. This was my thought when I read about video analytics-"what could be more interesting than this?". Each and every time I asked the above question whenever the keywords are changed. Soon, you need to identify the research problem or you need to prepare a better answer for your guide on why are you late?

• Overnight identification of research problem

 "Scholar: 💡  Professor!  I Identified the research problem
Professor: hmmm, right now you are in the third year of your Ph.D.,  Let me see"

        Most of the time, I heard from my fellow scholars during my Ph.D. journey that they found difficulty in identifying a research problem, and often research problems are changed in the middle of nowhere. This happened due to the overnight identification of the research problem. For example, each and every time when we see our guide we make our mind very positive by saying "Tomorrow, definitely I will report my research problem" and one day we will report at the verge of the deadline by reading a few works of literature. This delay may happen due to the hopping of interests, still hovering over broad topics, or our procrastination. 

Narrow Down your Research Problems in Hyperspectral (HS) Remote Sensing

       In hyperspectral remote sensing, with a simple grasp of paradigms and scopes of hyperspectral remote sensing, you can narrow down the research problems easily. There are three different paradigms in HS remote sensing

1. Pixel-wise classification
2. Sub-pixel-wise classification
3. Object-based classification

1. Pixel-wise classification

        The homogeneity spatial area considers that each pixel contains only pure pixel, then the classification of the class labels for the particular scene will be classified and it is referred to as a pixel-wise classification. In past decades, the traditional way of pixel-wise classification is widely used by the research community, to classify particular class labels.

2. Sub-pixel-wise classification

Figure 1: Linear and Nonlinear unmixing (from Vidhi Joshi et al., 2017)

        In recent decades, the researchers are focused on the sub-pixel-wise classification due to the intimate mixtures present in one pixel will affect the overall performance of the accuracy. This sub-pixel-wise classification is also referred to as spectral unmixing and it provides the detailed study of the fraction of the materials present in a pixel. Figure 1 shows that there is a presence of the three different components in one pixel. In spectral unmixing, there are two different unmixing approaches that are used inherently.

Linear unmixing and nonlinear unmixing approaches are based on the macroscopic and a microscopic level of intimate mixtures in a pixel respectively are shown in figure 1.

3. Object-based classification

At present, object-based classifications are in trend to classify the spectral information effectively. Here, when compared with the other two paradigms the geographical objects are considered as objects instead of pixel values. The generic framework of this paradigm contains the sequence of image segmentation, spatial, textural, and contextual information. This paradigm depends on the knowledge-based function which helps to classify the feature based on the rule. The hyperspectral image objects are grouped as per similar spatial or spectral properties.

Scopes

        The scopes I have considered for this post are based on the pixel-wise classification and sub-pixel-wise classification

Figure 2: Scopes in HS remote sensing with different paradigms

     Figure 2 shows the scopes of HS remote sensing with different paradigms. Pixel-wise classification plays a vital role in urban and rural area classification, agriculture cropping areas, exploration, and detection of the forest land cover areas and species, oceanography, etc. The sub-pixel-wise classification allows detecting the maturity stages of the fruits and vegetables, crop soil residue, etc., as well as in addition, to identify abundances of chemical properties present in our planets, mineralogy, etc.

Conclusion

    Once we know the destination, any hurdle in the path is not a problem. Likewise, we need to choose the right research problem to carry out our Ph.D. journey successfully. In this post for hyperspectral remote sensing, we can easily narrow down our research problems based on the different paradigms and with the scope of the study. I have done my Ph.D. research work in both pixel-wise and sub-pixel-wise classification. In real-time, we cannot expect homogeneity (Pure Pixel) in the HS dataset; there have always been intimate mixtures (nonlinear) present in a pixel. I suggest Nonlinear Sub-Pixel-Wise classification which improves the accuracy of each class label by employing the Artificial Intelligence models. Henceforth, you will be getting more information on Artificial Intelligence models and HS datasets. Please share your thoughts and comments about this post