Abstract/Summary

In using the convolutional neural network (CNN) for classification, there is a set of hyperparameters available for the configuration purpose. This study aims to validate the effectiveness of the CNN architecture, i.e. AlexNet on land cover classification, based on four remotely sensed land-use land-cover (LULC) datasets. In addition to the evaluation of the impact of a range of parameters in the evaluation tests, the influence of a set of hyperparameters on the classification performance will be assessed. The parameters include the epoch values, batch size, convolutional filter size and input image size. Thus, a set of experiments were conducted to specify the effectiveness of the selected parameters. We first explore the number of epochs under several selected batch size values. The impact of the first layer kernel size of the convolutional filters also was evaluated. Moreover, testing assorted sizes of the input images provided insight of the influence of the size of the convolutional filters and the image sizes. To generalize the validation, four remote sensing datasets, AID, RDS, UCMerced and RSCCN7, which have different land coverage and are publicly available, were used in the experiments. These datasets have a wide diversity of input data, such as the number of classes, amount of labelled data and texture patterns. A specifically-designed, interactive, deep-learning GPU training platform for image classification (Nvidia Digit) was employed in the experiments. It has shown efficiency in both training, evaluation and testing. These results provide opportunities toward better classification performance in various applications, such as hyperspectral multi-temporal agricultural LULC.