#TITLE_ALTERNATIVE#

• Main Author: Cakra Satya (NIM 233 97 004), Octavianus
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/8881
• Institution: Institut Teknologi Bandung

Abstract : <br /> <br /> <br /> This research is to design 4 interpolation technique using artificial neural network (ANN) that can be used for Digital Terrain Model (DTM) that is : Direct-Global ANN teenique, Direct-Local ANN technique, Scan-Global ANN technique and Scan-Local ANN technique. <br /> <br /> <br /> Direct-Global ANN technique can do 3 dimension interpolation with global data training, while Direct-Local ANN technique do similar to the technique above but the data training held by region around the test point. These two techniques use back propagation learning rule with Levenberg Marquardt algorithm, and can be applied to modify Irregular DIM into Grid DTM. <br /> <br /> <br /> Scan-Global ANN technique does 3 dimension interpolasion by double 2 dimension scanning with global data training, while Scan-Local ANN techniques by single 2 dimension scanning with data training held around the test point. These two techniques use Radial Basis Function Network (RBFN). Scan-Global ANN technique can be applied to modify Grid DTM into narrower Grid DTM, while Scan-Local DTM can be applied to change Profile DIM into Grid DTM. <br /> <br /> <br /> ANN learning and training is applied to DTM taken from the sintetic model and a model derived from real terrain data. The shape of sintetic model are cubic, half ball, conic, angle plane and combination. The model derived from real terrain data consists of : Region 1, Region 2 and Region 3. Their maximum height difference are 5, 10 and 20 meter respectively. <br /> <br /> <br /> Interpolation yield for Irregular DTM using Direct-Global and Direct Local ANN technique is better than that reached by Least Square method and Moving Surface method. Direct-Global ANN technique give better yield for the linear terrain such Region 2. On the other hand, Direct-Local ANN technique is the best for the terrain with break lines such Region 3. <br /> <br /> <br /> The performance of Scan-Local ANN technique is better than Scan-Global ANN technique. But these two ANN techniques do not give better yield than Linear and Cubic interpolation method. The succesful percentage is just around 50 %. With the same amount of training point, in several cases, Scan-Local ANN technique applied to Profile DTM can do better interpolastion than other methods. <br />