Asian Research Thesis Index

Abstract

The main purpose of this research is to build up a system, based on standard or objective parameters rather than non-standard or subjective parameters, which are already being employed by researchers, for the crop classification and crop growth analyses. The research is divided into two portions; the first part deals with the crop classification, whereas the second part is concerned with the crop growth analysis. For this purpose two types of datasets have been used; radiometric data and photographic data. In portion 1 radiometric data is acquired by using a handheld crop scan device ‗MSR5‘, in the form of five spectral bands, from 450nm to 1750nm, with five types of wavelength blue, green, red, infrared and far-infrared, whereas, the photographic data is obtained by a digital camera with14.1Mpixels resolution To meet the objectives a system has been developed and employed on two types of data; (a) test data and (b) experimental data. Both types of data (radiographic and photographic) are classified by using ANN classifier. In test data, five land classes are differentiated by this system. Photographic images of the same five types of land classification (as radiometric data) are used to extract following five types of 77 statistical textural features, which may be grouped as; first order (histogram) features, second order (GLCM) features, higher order (GLRM) features, autoregressive features, and gradient matrix based features are calculated from ROI (32x32),(64x64),(128x128),(256x256) and (512x512) by using MaZda software. The most relevant features for each size of ROI are selected by three approaches; Fisher‘s Co-efficient, Probability of Error plus Average Correlation Co-efficient, and Mutual Information Co-efficient. In this way the most relevant 10 features were selected by each method. We receive very poor results when data analysis capability is verified on the basis of 10 features are selected by each method for each size of ROI except (512x512), by three multivariate techniques; PCA, LDA, and NDA available in ‗B11‘, software integrated with MaZda. To improve the results, a set of 20 features is obtained by merging the features selected by each approach. An excellent clustering result with accuracy of 91.9% received, when data of these 20 features extracted from ROI (512x512) was deployed to NDA projection space. By using supervised classification approach, artificial neural network (ANN) the system is trained and tested on the basis of 70% and 30% of input data respectively. We received an accuracy of 100% and 91.33% in training and testing phase respectively. Similarly in radiometric data 250 data instances are taken for five different types of land (50 data instances for each type of land), for training purpose 40 data instances of each land type is used. Total 200 samples out of (250) are used to train the data system. Testing is performed on 50 samples (10 samples from each land type) and 96.40% accuracy result is obtained for radiometric data. On the basis of test data analysis, it is concluded that the proposed system produces the best result for large ROI window size when a combined set of features is deployed in NDA projection space. The photographic experimental data (five different types of crop) is analyzed under these settings. To check the system routine two disjoint sets of data with 70/30 ratio for training and testing respectively are developed ANN classifier available in B11 software under n-class training and testing option. is checked for the settings to which NDA has shown the best performance, Results show that the system training accuracy increases by number of neurons in input layer, and testing accuracy processes up to certain configuration. The best training accuracy of 85.17 and testing accuracy is 81.25% with 7 input layers at learning rate 0.35.For radiometric data ANN is trained and tested. For this purpose 400 scans data (80 scans from each class) is used to train the classifier and the remaining 100 scans data (20 scans from each class) was employed to test the classifier. We received an average accuracy of 94.50% during training and 96.00% accuracy in testing phase. In second part, which is concerned with the crop growth analysis, field data is acquired at different six stages by using crop scan MSR 5 (for radiometric data ) and a Photographic data was acquired by a digital camera mounted at a height by which approximately five square feet area is imaged in each photo. For every stage fifty images of photographic data are acquired from different regions of the crop field and the radiometric data is acquired at the altitude of 10 feet from the ground level. (This way the device scans an area of 5 square feet for each scan). For each stage approximately hundred scans are acquired by the said device from consecutive areas, the crop growth is assessed on the basis of reflectance values of five bands acquired by the devices at different stages. It is also observed that the same wave lengths (IR and FIR) are very helpful for the assessment of crop growth. For the growth assessment, in this study, we explore the changes in canopy spectrum feature of wheat. Reflectance patterns during the growing season expose a large amount of information about the changes in the visible and near-infrared (NIR) wavelengths. There is a rapid raise in the NIR values as soon as the crop develops while the changes in the visible wavebands adjust more slowly. Throughout the growing season the NIR wavelengths are more active than the visible wavelengths. This is due in part the reflectance values for bare soil being closer to the visible reflectance than NIR values. We bring to a close that the presence of infrared and far-infrared wavelengths makes the radiometric data more inventive for classification/differentiation as compared to photographic data.

Add/Update Thesis