AN IMPROVED ACCURACY PREPOSITION FOR VEGETATION CLASSIFICATION ON SATELLITE DATA AND A WEB BASED VEGETATION REAL TIME HEALTH SYSTEM

Abstract

According to the studies conducted by the World Bank, the food crisis is going to be a prime issue in next few years due to rising world population. Therefore, to bring sustainability agriculture sectors at every country in this globe is going through a paradigm shift by remote sensing tools for assessing the vegetation conditions. Data acquired by remote sensing tools like Satellites, UAVs, spectro-radiometers and smart phones reduce the human labour to estimate the spatial and temporal trajectory and also the health of vegetation at any particular region. However, processing of such data and extracting significant information from them require sufficient technical expertise. To make this process easier, recently web developers are developing paid subscription web portals having this inherent satellite data processing capability and to display the values of the vegetation indices (VIs). However, mostly these web pages require paid subscription, downloading and storing of large sized satellite images, and cannot do any other operation rather than giving only the raw VI values. Further, very few portals are available to process the UAV and smart phone images. Therefore, in order to provide effective algorithms to the web based vegetation monitoring system, the research works presented in this thesis are carried out. The main objectives are of two folds viz. designing new effective algorithms to process different remote sensing data and development of open access web portals to bring the advantage of some of these algorithms accessible to the end users. A short description about those research works are provided below. At first, a new conjugated dense CNN (CD-CNN) architecture with a new activation function named SL-ReLU is proposed for intelligent classification of multiple crops from RGB images captured by UAV. CD-CNN integrates data fusion and feature map extraction in conjunction with classification process. Initially a dense block architecture is proposed with a new activation function, called SL-ReLU, associated with the convolution operation to mitigate the chance of unbounded convolved output and gradient explosion. Dense block architecture concatenates all the previous layer features for determining the new features. This reduces the chance of losing important features due to deepening of the CNN module. Later, two dense blocks are conjugated with the help of a conversion block for obtaining better performance. Unlike traditional CNN, CD-CNN omits the use of fully connected layer and that reduces the chance of feature loss due to random weight initialization. The proposed CD-CNN achieves a strong distinguishing capability from several classes of crops. Raw UAV images of five different crops are captured from different parts of India and then small candidate crop regions are extracted from the raw images with the help of Arc GIS 10.3.1 software and then the candidate regions are fed to CD-CNN for proper training purpose. Experimental results show that the proposed module can achieve an accuracy of 96.2% for the concerned data. Afterwards, a study is conducted between freely available low-resolution satellite and high-resolution UAV data to evaluate their capability for approximating land cover types and areas. In this study, data captured from a UAV equipped with a Multispectral Mica Sense Red Edge camera used as ground-truth information to calibrate Sentinel-2 imagery. UAV-based NDVI allowed crop estimation at 10-cm pixel resolution by discriminating no-green vegetation pixels. The reflectance value and NDVI of the crops at different stages were derived from both UAV and Sentinel-2 images. The UAV Multispectral mapping method used in this study provided advanced information about the physical conditions of the study area (Roorkee) and improved land feature delineation. The result shows that UAV data produced more accurate reflectance values than Sentinel-2 imagery. Then to draw a comparison between three different remote sensing tools to derive the vegetation indices, an analysis between Landsat-8, UAV and field spectro-radiometer observed data is performed. The current study aims to utilize unmanned and multispectral satellite imaging in integration with spectro-radiometer for estimation of basic crop parameters during the growing season. The UAV DJI-Inspire 2 T650A UAV with Mica Sense Red Edge-M™ multispectral camera was used to acquire very high-resolution multispectral imagery in RGB, and Near NIR band, Landsat-8 satellite image of the same date is downloaded from the satellite data distribution platform (earth explorer) of NASA and a detailed survey is conducted SVC HR-1024i spectro radiometer for ground data collection for the same date. NDVI values obtained from UAV datasets are compared with the Landsat-8 derived NDVI values and reference spectro- radiometer observations for specific crop cover at different sampling sites. This helps to demonstrate the utility and effectiveness of UAV multispectral camera in contrast to Landsat-8 satellite imagery for the observation of crops. As, the performance of spectro-radiometer is superior among three tools, so later using this tool the impact of polluted urban environment on the leaf pigments and water content in a tree are investigated. The current state of art examines effects of air pollution on four major biochemical perimeters of tree health i.e. chlorophyll content, water content, carotenoid content and anthocyanin content. A spectro-radiometer and remote sensing indices integrated approach is used to evaluate the impact of deteriorating air quality on mango tree planted in and around the Indian Institute of Technology Roorkee, campus. Four different Vegetation Indices content (Normalized Difference Vegetation Indices, modified normalized difference vegetation index, simple ratio and modified simple ratio) are used for estimate chlorophyll. Five different Water indices (Water index, Normalized water indices-1, Normalized water indices-2, Normalized water indices-3 and Normalized water indices-4) are used for estimating water content. Five different photochemical information indices (Carotenoid concentration index, photochemical reflectance index, plant senescencing reflectance index, and Carotenoid concentration index) are used for enumerate carotenoid content. Three different Anthocyanin Reflectance Indices (Modified Anthocyanin Content Index, Anthocyanin Reflectance Index and Modified Anthocyanin Reflectance Index) are used for determining anthocyanin content. Thereafter, an attention dense learning (ADL) mechanism is proposed by merging mixed sigmoid attention learning with the basic dense learning process of deep CNN. The basic dense learning process derives new features at higher layer considering all lower layer features and that provides fast and efficient training process. Further, the attention learning process amplifies the learning ability of the dense block by discriminating the meaningful lesion portions of the images from the background areas. Other than adding an extra layer for attention learning, in the proposed ADL block the output features from higher layer dense learning are used as an attention mask to the lower layers. For an effective and fast classification process, five ADL blocks are stacked to build a new CNN architecture named DADCNN-5 for obtaining classification robustness and higher testing accuracy. The efficacy of the DADCNN-5 model is checked after performing stringent experiments on a new real world plant leaf database, created by the authors. The new leaf database contains 10,851 real-world RGB leaf images of 17 plant species for classifying their 44 distinguished health conditions. Further, the robustness of the DADCNN-5 is established after experimenting with augmented and noise contaminated images of the practical database. Again, to reduce the chance of accuracy decrement due to erroneous choice of the training hyperparameters, Opposition-based Symbiotic Organisms Search (OSOS) algorithm is implemented for optimizing the values of learning rate and momentum during the training process. At last, three open access web portals are designed to assess the vegetation health using satellite, UAV and smart phone images. First Web-GIS portal process the Sentinel-2 and MODIS data to calculate the NDVI values for India. Direct access to the satellite website, time to time update of the data and availability of different facilities like NDVI calculation on a particular date, temporal change in NDVI values etc. make this portal suitable. Time series histogram plotting and video graphic representation of the results make this portal user friendly. Second web portal uses the OSOS-DADCNN-5 network to predict the different leaf diseases of several plants species. The third web portal is able to process the multispectral UAV images and calculate twelve different vegetation indices to inform the user about the vegetation density at any particular region. Colour map representation for each calculated VI makes this portal attractive. The presented works in this thesis are likely to contribute significantly to the area of web based vegetation monitoring using the remote sensing data. The different developed techniques will be particularly useful for assessing the health conditions of forest area, crop lands, and trees for estimating net yield production. Some future research suggestions on observations and simulations in this research area are proposed at the end of the thesis for the benefit of potential researchers.