Classification of Very High Resolution Aerial Photos Using Spectral-Spatial Convolutional Neural Networks

Abstract

Classification of aerial photographs relying purely on spectral content is a challenging topic in remote sensing. A convolutional neural network (CNN) was developed to classify aerial photographs into seven land cover classes such as building, grassland, dense vegetation, waterbody, barren land, road, and shadow. The classifier utilized spectral and spatial contents of the data to maximize the accuracy of the classification process. CNN was trained from scratch with manually created ground truth samples. The architecture of the network comprised of a single convolution layer of 32 filters and a kernel size of 3 × 3, pooling size of 2 × 2, batch normalization, dropout, and a dense layer with Softmax activation. The design of the architecture and its hyperparameters were selected via sensitivity analysis and validation accuracy. The results showed that the proposed model could be effective for classifying the aerial photographs. The overall accuracy and Kappa coefficient of the best model were 0.973 and 0.967, respectively. In addition, the sensitivity analysis suggested that the use of dropout and batch normalization technique in CNN is essential to improve the generalization performance of the model. The CNN model without the techniques above achieved the worse performance, with an overall accuracy and Kappa of 0.932 and 0.922, respectively. This research shows that CNN-based models are robust for land cover classification using aerial photographs. However, the architecture and hyperparameters of these models should be carefully selected and optimized.

Land Cover Classification from fused DSM and UAV Images Using Convolutional Neural Networks

“…The two datasets are: (i) the RGB image and (ii) the fusion of RGB and DSM data. Patch-level analysis is often used with deep learning methods, especially CNN in order to overcome challenges posted by speckle noise and segmentation optimization which problems associated with pixel-level and object-level feature extraction [18]. In a patch-level (also known as tile-based) analysis, images are divided into a grid of tiles of m × m and then each patch is separately analyzed.…”

“…Very-high resolution aerial images were classified using a CNN in [29], which has been shown to be effective for the extraction of specific objects such as cars. In another study [18], the capability of CNN to classify aerial photos (with 10 cm resolution) was examined and verified using medium-scale datasets.…”

“…CNN is an artificial neural network that uses the local receptive field Remote Sens. 2019, 11, 1461 6 of 18 and shared weights [18]. This deep learning method has the ability to automatically learn hierarchically contextual features from the input image and this architecture provide a suitable platform to analyze very-high-resolution images [29].…”

“…The architecture of CNN is based on three series of operations (layers): 2D convolution layer, pooling layer (subsampling), and fully connected layer [36]. Additionally, some deeper structures can be used to improve CNNs generalization such as nonlinear activations, regularization (dropout and batch normalization), stochastic pooling, and so on [18,38].…”

“…Specifically, we investigated the effectiveness of using a single feature modality (RGB data only) versus the fusion of RGB and DSM data. The classification algorithm used to compare the two datasets is a convolutional neural network (CNN), which is a deep-learning technique that hierarchically learns representations from training data, regulates and shares the weight with respect to the training data, generalizes, optimizes, and reduces the parameters with the higher ability to discriminate and extract features, automatically [18,19]. The specific land cover classifications that we considered were: (i) bare land, (ii) buildings, (iii) dense vegetation/trees, (iv) grassland, (v) paved roads, (vi) shadows, and (vii) water bodies.…”