Deep learning is a fast-growing area of machine learning research that can be applied to image analysis and feature extraction. It involves Neural Networks (NNs) with multiple layers that can extract feature representations, often unsupervised and exclusively from data. Internet companies such as Google, Baidu, Microsoft and Facebook make extensive use of deep learning for image analysis tasks such as image indexing, segmentation and object/face detection and recognition. Deep learning techniques have also been successfully applied to many data analysis tasks including handwriting, video, speech recognition and natural language processing and have been applied across multiple domains including vehicle automation, medical imaging, drug discovery, genomics as well as remote sensing and the geosciences.
In this research study, the contributions that new deep learning models can make to automated classification are investigated, using satellite and photographic imagery to identify plants, crops, vegetation as well as land cover. The project will focus mainly on developing new techniques to discriminate broad land cover classes in Australia using deep learning models. Results will be compared against benchmark datasets and with alternate machine learning techniques for classification such as Random Forest decision trees.