Coarse-to-Fine Satellite Images Change Detection Framework via Boundary-Aware Attentive Network

Timely and accurate change detection on satellite images by using computer vision techniques has been attracting lots of research efforts in recent years. Existing approaches based on deep learning frameworks have achieved good performance for the task of change detection on satellite images. However, under the scenario of disjoint changed areas in various shapes on land surface, existing methods still have shortcomings in detecting all changed areas correctly and representing the changed areas boundary. To deal with these problems, we design a coarse-to-fine detection framework via a boundary-aware attentive network with a hybrid loss to detect the change in high resolution satellite images. Specifically, we first perform an attention guided encoder-decoder subnet to obtain the coarse change map of the bi-temporal image pairs, and then apply residual learning to obtain the refined change map. We also propose a hybrid loss to provide the supervision from pixel, patch, and map levels. Comprehensive experiments are conducted on two benchmark datasets: LEBEDEV and SZTAKI to verify the effectiveness of the proposed method and the experimental results show that our model achieves state-of-the-art performance.

Accuracy Assessments of Local and Global Forest Change Data to Estimate Annual Disturbances in Temperate Forests

Forest disturbances are generally estimated using globally available forest change maps or locally calibrated disturbance maps. The choice of disturbance map depends on the trade-offs among the detection accuracy, processing time, and expert knowledge. However, the accuracy differences between global and local maps have still not been fully investigated; therefore, their optimal use for estimating forest disturbances has not been clarified. This study assesses the annual forest disturbance detection of an available Global Forest Change map and a local disturbance map based on a Landsat temporal segmentation algorithm in areas dominated by harvest disturbances. We assess the forest disturbance detection accuracies based on two reference datasets in each year. We also use a polygon-based assessment to investigate the thematic accuracy based on each disturbance patch. As a result, we found that the producer’s and user’s accuracies of disturbances in the Global Forest Change map were 30.1–76.8% and 50.5–90.2%, respectively, for 2001–2017, which corresponded to 78.3–92.5% and 88.8–97.1%, respectively in the local disturbance map. These values indicate that the local disturbance map achieved more stable and higher accuracies. The polygon-based assessment showed that larger disturbances were likely to be accurately detected in both maps; however, more small-scale disturbances were at least partially detected by the Global Forest Change map with a higher commission error. Overall, the local disturbance map had higher forest disturbance detection accuracies. However, for forest disturbances larger than 3 ha, the Global Forest Change map achieved comparable accuracies. In conclusion, the Global Forest Change map can be used to detect larger forest disturbances, but it should be used cautiously because of the substantial commission error for small-scale disturbances and yearly variations in estimated areas and accuracies.

A Classified Adversarial Network for Multi-Spectral Remote Sensing Image Change Detection

Adversarial training has demonstrated advanced capabilities for generating image models. In this paper, we propose a deep neural network, named a classified adversarial network (CAN), for multi-spectral image change detection. This network is based on generative adversarial networks (GANs). The generator captures the distribution of the bitemporal multi-spectral image data and transforms it into change detection results, and these change detection results (as the fake data) are input into the discriminator to train the discriminator. The results obtained by pre-classification are also input into the discriminator as the real data. The adversarial training can facilitate the generator learning the transformation from a bitemporal image to a change map. When the generator is trained well, the generator has the ability to generate the final result. The bitemporal multi-spectral images are input into the generator, and then the final change detection results are obtained from the generator. The proposed method is completely unsupervised, and we only need to input the preprocessed data that were obtained from the pre-classification and training sample selection. Through adversarial training, the generator can better learn the relationship between the bitemporal multi-spectral image data and the corresponding labels. Finally, the well-trained generator can be applied to process the raw bitemporal multi-spectral images to obtain the final change map (CM). The effectiveness and robustness of the proposed method were verified by the experimental results on the real high-resolution multi-spectral image data sets.

ANALISIS PERUBAHAN PENGGUNAAN LAHAN MENGGUNAKAN SISTEM INFORMASI GEOGRAFIS (SIG) DI GAMPONG ALUE NAGA KECAMATAN SYIAH KUALA TAHUN 2004-2019

Gampong Alue Naga merupakan salah satu gampong di Kecamatan Syiah Kuala Kota Banda Aceh dan berada di kawasan pesisir yang berbatasan langsung dengan Selat Malaka. Gampong Alue Naga dipisahkan oleh Sungai Lamyong yang berada ditengah gampong. Pasca tsunami penggunaan lahan di Gampong Alue Naga mengalami perubahan yang cukup besar. Penelitian ini mengangkat masalah tentang perubahan penggunaan lahan yang terjadi Gampong Alue Naga Kecamatan Syiah Kuala. Tujuan penelitian ini adalah untuk mengetahui seberapa besar perubahan penggunaan lahan yang terjadi di Gampong Alue Naga menggunakan Sistem Informasi Geografis (SIG) menggunakan aplikasi ArcGIS pada citra temporal Gampong Alue Naga. Adapun citra temporal yang digunakan yaitu citra tahun 2004, 2009, dan 2019. Untuk melihat perubahan penggunaan lahan yang ada di Gampong Alue Naga maka peta penggunaan lahan di overlay hingga mendapatkan peta yang baru yaitu peta perubahan penggunaan lahan. Hasil menunjukkan bahwa, perubahan lahan yang terjadi di Gampong Alue Naga Kecamatan Syiah Kuala cukup signifikan pada beberapa lahan seperti lahan terbangun, kebun kelapa, dan lahan tambak. ABSTRACT Gampong Alue Naga is one of the villages in the Syiah Kuala District of Banda Aceh City and is located in a coastal area directly adjacent to the Selat Malaka. Gampong Alue Naga is separated by the Lamyong River in the center of the village. After the tsunami the land use in Gampong Alue Naga has experienced significant changes. This study raises the problem of land use changes that occur in Gampong Alue Naga, Syiah Kuala District. The purpose of this study was to determine how much land use change occurred in the Gampong Alue Naga using a Geographic Information System (GIS) using the ArcGIS application on the temporal image of the Gampong Alue Naga. The temporal image used is the image in 2004, 2009 and 2019. To see changes in land use in the Gampong Alue Naga, the land use map is overlaid to get a new map, namely the land use change map. The results show that land changes that occurred in Gampong Alue Naga, Syiah Kuala District were quite significant in some lands such as built up land, coconut plantations, and pond areas.

EARTHQUAKE DAMAGE DETECTION USING SATELLITE IMAGES (CASE STUDY: SARPOL-ZAHAB EARTHQUAKE)

After the earthquake in Sarpol-e Zahab city, many people were killed or wounded and many buildings were destroyed. After such a destructive event, it is of great interest to efficiently identify the magnitude and the extent of the damaged areas. Remote sensing is an excellent technology for this purpose. Usually, a higher success rate can be achieved when both pre and post-event data, especially multi-view data, are used. Whereas when only post-event data are available, the detection is usually limited to the block level unless VHR images of a resolution of 0.5 m or higher are involved.The available dataset consists of one Pleiades-1 satellite optical image (post-event), two Sentinel-2 satellite images (pre&post event).After classification of the sentinel images(pre&post event) and preparation change maps by means of SVM and the neural network classification methods, two change maps will be provided. Then, A reference change map is prepared with ROIs. For this purpose, on the Pleiades-1 image (after the earthquake), ROIs in two categories “change” and “no change” are defined. In the last step, using the confusion matrix, two change maps from the Sentinel image are compared to the reference image, and the results are analyzed. The producer’s accuracy for detecting the collapsed buildings in the SVM classification method was found to be 78.34% and for the neural network classification was found to be 72.43%. The results show that the change map of the pre- and post-earthquake medium-resolution satellite images such as Sentinel-2 can reveal the collapsed buildings caused by the earthquake successfully.

Fusion of Change Vector Analysis in Posterior Probability Space and Postclassification Comparison for Change Detection from Multispectral Remote Sensing Data

Postclassification Comparison (PCC) has been widely used as a change-detection method. The PCC algorithm is straightforward and easily applicable to all satellite images, regardless of whether they are acquired from the same sensor or in the same environmental conditions. However, PCC is prone to cumulative error, which results from classification errors. Alternatively, Change Vector Analysis in Posterior Probability Space (CVAPS), which interprets change based on comparing the posterior probability vectors of a pixel, can alleviate the classification error accumulation present in PCC. CVAPS identifies the type of change based on the direction of a change vector. However, a change vector can be translated to a new position within the feature space; consequently, it is not inconceivable that identical measures of direction may be used by CVAPS to describe multiple types of change. Our proposed method identifies land-cover transitions by using a fusion of CVAPS and PCC. In the proposed algorithm, contrary to CVAPS, a threshold does not need to be specified in order to extract change. Moreover, the proposed method uses a Random Forest as a trainable fusion method in order to obtain a change map directly in a feature space which is obtained from CVAPS and PCC. In other words, there is no need to specify a threshold to obtain a change map through the CVAPS method and then combine it with the change map obtained from the PCC method. This is an advantage over other change-detection methods focused on fusing multiple change-detection approaches. In addition, the proposed method identifies different types of land-cover transitions, based on the fusion of CVAPS and PCC, to improve the results of change-type determination. The proposed method is applied to images acquired by Landsat and Quickbird. The resultant maps confirm the utility of the proposed method as a change-detection/labeling tool. For example, the new method has an overall accuracy and a kappa coefficient relative improvement of 7% and 9%, respectively, on average, over CVAPS and PCC in determining different types of change.

Building Instance Change Detection from Large-Scale Aerial Images using Convolutional Neural Networks and Simulated Samples

We present a novel convolutional neural network (CNN)-based change detection framework for locating changed building instances as well as changed building pixels from very high resolution (VHR) aerial images. The distinctive advantage of the framework is the self-training ability, which is highly important in deep-learning-based change detection in practice, as high-quality samples of changes are always lacking for training a successful deep learning model. The framework consists two parts: a building extraction network to produce a binary building map and a building change detection network to produce a building change map. The building extraction network is implemented with two widely used structures: a Mask R-CNN for object-based instance segmentation, and a multi-scale full convolutional network for pixel-based semantic segmentation. The building change detection network takes bi-temporal building maps produced from the building extraction network as input and outputs a building change map at the object and pixel levels. By simulating arbitrary building changes and various building parallaxes in the binary building map, the building change detection network is well trained without real-life samples. This greatly lowers the requirements of labeled changed buildings, and guarantees the algorithm’s robustness to registration errors caused by parallaxes. To evaluate the proposed method, we chose a wide range of urban areas from an open-source dataset as training and testing areas, and both pixel-based and object-based model evaluation measures were used. Experiments demonstrated our approach was vastly superior: without using any real change samples, it reached 63% average precision (AP) at the object (building instance) level. In contrast, with adequate training samples, other methods—including the most recent CNN-based and generative adversarial network (GAN)-based ones—have only reached 25% AP in their best cases.