Dynamic Multicontext Segmentation of Remote Sensing Images Based on Convolutional Networks

Abstract

Semantic segmentation requires methods capable of learning high-level features while dealing with large volume of data. Towards such goal, Convolutional Networks can learn specific and adaptable features based on the data. However, these networks are not capable of processing a whole remote sensing image, given its huge size. To overcome such limitation, the image is processed using fixed size patches. The definition of the input patch size is usually performed empirically (evaluating several sizes) or imposed (by network constraint). Both strategies suffer from drawbacks and could not lead to the best patch size. To alleviate this problem, several works exploited multi-context information by combining networks or layers. This process increases the number of parameters resulting in a more difficult model to train. In this work, we propose a novel technique to perform semantic segmentation of remote sensing images that exploits a multi-context paradigm without increasing the number of parameters while defining, in training time, the best patch size. The main idea is to train a dilated network with distinct patch sizes, allowing it to capture multi-context characteristics from heterogeneous contexts. While processing these varying patches, the network provides a score for each patch size, helping in the definition of the best size for the current scenario. A systematic evaluation of the proposed algorithm is conducted using four high-resolution remote sensing datasets with very distinct properties. Our results show that the proposed algorithm provides improvements in pixelwise classification accuracy when compared to state-of-the-art methods.

A Novel Deep Structure U-Net for Sea-Land Segmentation in Remote Sensing Images

“…Several features were firstly extracted from the entire image and later used to learn a robust sea-land classifier which converts the segment issues into a binary classification problem. In [24], the authors developed a technique to perform semantic segmentation for remote sensing images that uses a multi-scale model without increasing the number of parameters via optimization. The key idea is to train a dilated network with different patch sizes, to gain multi-scale features from heterogeneous contexts.…”

“…It is notable that almost half of the computation cost belongs to the last two blocks which have high resolutions, and needs more time to construct them. To analyze the effect of the proposed DenseNet, we use the evaluation metrics proposed in [24]. In Fig.…”

“…DeepUNet has the overall accuracy of 96.5% and the error rate of 7%. The Multi-Scale Segmentation model [24] has 96.3% accuracy and 9% error rate. The FusionNet, U-Net, and SegNet have the accuracy of 96%, 95% and 94% with the error rates of 9%, 12%, and 14% respectively.…”

“…The FusionNet, U-Net, and SegNet have the accuracy of 96%, 95% and 94% with the error rates of 9%, 12%, and 14% respectively. To assess the overall performance of the proposed RDU-Net, on the whole dataset, we compare its results with those of the eight deep neural network models (U-Net [9], SegNet [13], ResNet [11], FusionNet [15], Basaeed et al [17], Nogueira et al [24], DenseNet [12] and DeepUNet [5]) and two classical machine learning models (SVM and Random Forest). The results are illustrated in Table VII.…”

“…We intend to reduce the distance between the layers during the backpropagation that has a significant role in reducing the computation costs. Meanwhile, short-range connections and identity mapping [44] are combined to build a more effective network than the current approaches [5], [15], where identity mapping is a fundamental aspect of learning in deep networks and has significant effects in the training of the networks [24].…”