Recently, multicategory object detection in high-resolution remote sensing images is still a challenge. First, objects with significant scale differences exist in one scene simultaneously, so it is generally difficult for the detectors to balance the detection performance of large and small objects. Second, because of the complex background and the objects’ densely distributed characteristics in the remote sensing images, the extracted features usually have noise and blurred boundaries, which interfere with the detection performance of the object detectors. With this observation, we propose an end-to-end scale-aware network called feature split–merge–enhancement network (SME-Net) for remote sensing object detection, composed of the feature split-and-merge (FSM) module, the offset-error rectification (OER) module, and the object saliency enhancement (OSE) strategy. FSM eliminates salient information of large objects to highlight the features of small objects in the shallow feature maps. It also transmits the effective detailed features of large objects to the deep feature maps, alleviating feature confusion between multiscale objects. OER corrects the inconsistency of the features spatial layout among the multilayer feature maps by the proposed offset loss, so as to achieve supervised elimination and transmission in FSM. OSE enhances the features of interests and suppresses the background information by the proposed membership function, thus preventing false detection and missed detection caused by noise and blurred boundaries. The effectiveness of the proposed algorithm has been verified on multiple datasets. Our code is available at: https://github.com/Momuli/SMENet.git

Deep learning has obtained state-of-the-art results in a variety of computer vision tasks and has also been used to solve SAR image classification problems. Deep learning algorithms typically require a large amount of training data to achieve high accuracy. In contrast, the size of SAR image datasets is often comparatively limited. Therefore, this paper proposes a novel method, deep memory convolution neural networks (M-Net), to alleviate the problem of overfitting caused by insufficient SAR image samples. Based on the convolutional neural networks (CNN), M-Net adds an information recorder to remember and store samples’ spatial features, and then it uses spatial similarity information of the recorded features to predict unknown sample labels. M-Net's use of this information recorder may cause difficulties for convergence if conventional CNN training methods were directly used to train M-Net. To overcome this problem, we propose a transfer parameter technique to train M-Net in two steps. The first step is to train a CNN, which has the same structure as the part of CNN incorporated in M-Net, to obtain initial training parameters. The second step applies the initialized parameters to M-Net and then trains the entire M-Net. This two-step training approach helps us to overcome the nonconvergence issue, and also reduces training time. We evaluate M-Net using the public benchmark MSTAR dataset, and achieve higher accuracy than several other well-known SAR image classification algorithms.

To detect changed areas in multitemporal polarimetric synthetic aperture radar (SAR) images, this paper presents a novel version of convolutional neural network (CNN), which is named local restricted CNN (LRCNN). CNN with only convolutional layers is employed for change detection first, and then LRCNN is formed by imposing a spatial constraint called local restriction on the output layer of CNN. In the training of CNN/LRCNN, the polarimetric property of SAR image is fully used instead of manual labeled pixels. As a preparation, a similarity measure for polarimetric SAR data is proposed, and several layered difference images (LDIs) of polarimetric SAR images are produced. Next, the LDIs are transformed into discriminative enhanced LDIs (DELDIs). CNN/LRCNN is trained to model these DELDIs by a regression pretraining, and then a classification fine-tuning is conducted with some pseudolabeled pixels obtained from DELDIs. Finally, the change detection result showing changed areas is directly generated from the output of the trained CNN/LRCNN. The relation of LRCNN to the traditional way for change detection is also discussed to illustrate our method from an overall point of view. Tested on one simulated data set and two real data sets, the effectiveness of LRCNN is certified and it outperforms various traditional algorithms. In fact, the experimental results demonstrate that the proposed LRCNN for change detection not only recognizes different types of changed/unchanged data, but also ensures noise insensitivity without losing details in changed areas.

A new polarimetric synthetic aperture radar (PolSAR) images classification method based on multilayer autoencoders and superpixels is proposed in this paper. First, in order to explore the spatial relations between pixels in PolSAR data, the RGB image formed with Pauli decomposition is used to produce superpixels to integrate contextual information of neighborhood. Second, multilayer autoencoders network is used to learning the features used for distinguishing the multiple categories for each pixel, and a softmax regression is applied to produce the predicted probability distributions over all the classes of each pixel. Finally, the probability distributions is regarded as a new probabilistic metric and introduced to k-nearest neighbor to improve the accuracy of classification based on superpixels, which takes spatial relationship between pixels into consideration, and it is robust to speckle noise. The proposed method makes good use of the scattering characteristics in each pixel and spatial information of PolSAR data. Compared with other state-of-the-art methods, the results of proposed method show better agreement with the ground truth and significant improvement in classification accuracy and discriminability of small differences between different categories.

Supervised deep neural networks (DNNs) have been extensively used in diverse tasks. Generally, training such DNNs with superior performance requires a large amount of labeled data. However, it is time-consuming and expensive to manually label the data, especially for tasks in remote sensing, e.g., change detection. The situation motivates us to resort to the existing related images with labels, from which the concept of change can be adapted to new images. However, the distributions of the related labeled images (source domain) and unlabeled new images (target domain) are similar but not identical. It impedes a change detection model learned from source domains being well applied to the target domain. In this paper, we propose a transferred deep learning-based change detection framework to solve this problem. It consists of pretraining and fine-tuning stages. In the pretraining process, we propose two tasks to be learned simultaneously, namely, change detection for the source domain with labels and reconstruction of the unlabeled target data. The auxiliary task aims to reconstruct the difference image (DI) for the target domain. DI is an effective feature, such that the auxiliary task is of much relevance to change detection. The lower layers are shared between these two tasks in the training process. It mitigates the distribution discrepancy between the source and target domains and makes the concept of change from the source domain adapt to the target domain. In addition, we evaluate three modes of the U-net architecture to merge the information for a pair of patches. To fine-tune the change detection network (CDN) for the target domain, two strategies are exploited to select the pixels that have a high possibility of being correctly classified by an unsupervised approach. The proposed method demonstrates an excellent capacity for adapting the concept of change from the source domain to the target domain. It outperforms the state-of-the-art change detection methods via experimental results on real remote sensing data sets.