preview

Remote Sensing Image Scene Essay

Decent Essays
Open Document

Chapter 5 Result Analysis 5.1 Introduction In this chapter, we will discuss our experimental results. The aim of this experiment to find more accurate results for remote sensing image scene classification. In this experiment we have used NWPU-RESISC45 dataset. We applied some normal learning procedure and deep learning procedure in this experiment. In deep learning procedure we used two well-known deep learning named VGG16 and ResNet50. We, trained them from scratch as well as applied transfer learning on our dataset. We discussed about the procedures in the previous chapter. We will discuss the results, make comparison and eventually draw a conclusion on this chapter. We will also discuss about implementation environment and dataset. 5.2 …show more content…

Using different procedures our findings are given below in Table 5.1. Table 5.1: Experimental results of different methods. Method Accuracy (%) NABP 71.76% Centrist 76.5% BVW 76.8% VGG11 76.9% Model1 87.57% VGG16 (classifier) 91.9% VGG16(fine-tuning) 90.48% ResNet (classifier) 95.9% ResNet (fine-tune) 96.01% A comparison based on different findings using different procedures is given in Figure 5.1 Figure 5.1: Experimental results comparison for different methods on NWPU-RESISC45 dataset In the Figure 5.1, we can categorize procedures into three parts where one part is normal learning, second part is deep learning from scratch and anther is transfer learning. We found less accuracies using normal learning approach because normal learning procedures are not so robust. We found better accuracies using convolutional neural network training from scratch. We found our best results using transfer learning. The reason is that we are using a comparatively small dataset. When we train a model from scratch it needs to update whole parameters while training. As dataset is comparatively small, a large number of parameter get over fitted on these small data. When we are using transfer learning we do not interfere the generic feature extraction of a model as it learned before. We just train a limited number of parameters to fit into our data. So, the model does not get over fitted and produce a good performance. 5.5 Conclusion In this chapter we have

Get Access