Abstrakt: |
Rise in global wide population has created lot of demand across various fields and realm. As a result, many recent research mainly focuses on LCC (Land Cover Classification) which are associated with spatial distribution. Therefore, LCC plays an important role for government organizations, policy makers and farmers in order to enhance the process of decision making. In addition to that, LCC using remote sensing has possess various applications such as urban planning, precision agriculture. A massive volume of heterogeneous geographical images over a wide range of geographical areas associated with diversified imaging conditions generally cause photographic distortions, illumination change, and the scale variation, inaccuracy, inefficient due to implementation of ineffective algorithms in the existing studies that seriously decline the classification accuracy. Therefore, proposed study implemented deep learning techniques for overcoming these issues since it minimizes the computation time, makes the network converge much faster and reduces the over-fitting, moreover ResNet50 is light weight deep learning model which are fast to train when appropriately scaled for depth, width and input data resolution can provide comparable and even higher image classification accuracies. This is especially important in remote sensing where the volume of data is very large and increases constantly. Thereby gaining positive implications and delivering exceptional performances in land cover classification in the proposed study. Hence proposed study incorporated a D-CNN (Deep CNN) and ResNet50 for extracting the appropriate features from the pre-processed dataset. Once the data are extracted, dimensionality reduction takes by employing PCA (Principal Component Analysis) to rule out more number of irrelevant features. After employing PCA, classification of images takes place by implementing logistic weight updated hyper parameter tuned random forest method which classifies the extracted features. Finally, performance of the proposed model is evaluated using different performance metrics like accuracy, precision, recall and F1-score. Then the proposed method is further evaluated by comparing the proposed method with the existing methods for assessing the efficiency and efficacy of the proposed model. [ABSTRACT FROM AUTHOR] |