Yoriko Kazama

House damage assesment based on supervised learning method: case study on Haiti

Assessing the damage caused by natural disasters requires fast and reliable information. Satellite imagery, especially high-resolution imagery, is recognized as an important source for wide-range and immediate data acquisition. Disaster assessment using satellite imagery is required worldwide. To assess damage caused by an earthquake, house changes or landslides are detected by comparing images taken before and after the earthquake. We have developed a method that performs this comparison using vector data instead of raster data. The proposed method can detect house changes without having to rely on various image acquisition situations and shapes of house shadows. We also developed a houseposition detection method based on machine learning. It uses local features including not only pixel-by-pixel differences but also the shape information of the object area. The result of the house-position detection method indicates the likelihood of a house existing in that area, and changes to this likelihood between multi-temporal images indicate the damaged house area. We evaluated our method by testing it on two WorldView-2 panchromatic images taken before and after the 2010 earthquake in Haiti. The highly accurate results demonstrate the effectiveness of the proposed method.

Towards automation of building damage detection using WorldView-2 satellite image: the case of the Haiti earthquake

Information of disaster damage assessment is very significant to disaster mitigation, aid and post disaster redevelopment planning. Remotely sensed data, especially very high resolution image data from aircraft and satellite have been long recognized very essential and objective source for disaster mapping. However feature extraction from these data remains a very challenge task currently. In this paper, we present a method to extract building damage caused by earthquake from two pairs of Worldview-2 high resolution satellite image. Targeting at implementing a practically operational system, we develop a novel framework integrating semi-automatic building extraction with machine learning mechanism to maximize the automation level of system. We also present a rectilinear building model to deal with a wide variety of rooftops. Through the study case of Haiti earthquake, we demonstrate our method is highly effective for detecting building damage from high resolution satellite image.

Shallow water bathymetry correction using sea bottom classification with multispectral satellite imagery

Bathymetry at shallow water especially shallower than 15m is an important area for environmental monitoring and national defense. Because the depth of shallow water is changeable by the sediment deposition and the ocean waves, the periodic monitoring at shoe area is needed. Utilization of satellite images are well matched for widely and repeatedly monitoring at sea area. Sea bottom terrain model using by remote sensing data have been developed and these methods based on the radiative transfer model of the sun irradiance which is affected by the atmosphere, water, and sea bottom. We adopted that general method of the sea depth extraction to the satellite imagery, WorldView-2; which has very fine spatial resolution (50cm/pix) and eight bands at visible to near-infrared wavelengths. From high-spatial resolution satellite images, there is possibility to know the coral reefs and the rock area’s detail terrain model which offers important information for the amphibious landing. In addition, the WorldView-2 satellite sensor has the band at near the ultraviolet wavelength that is transmitted through the water. On the other hand, the previous study showed that the estimation error by the satellite imagery was related to the sea bottom materials such as sand, coral reef, sea alga, and rocks. Therefore, in this study, we focused on sea bottom materials, and tried to improve the depth estimation accuracy. First, we classified the sea bottom materials by the SVM method, which used the depth data acquired by multi-beam sonar as supervised data. Then correction values in the depth estimation equation were calculated applying the classification results. As a result, the classification accuracy of sea bottom materials was 93%, and the depth estimation error using the correction by the classification result was within 1.2m.

Ship detection leveraging deep neural networks in WorldView-2 images

Interpretation of high-resolution satellite images has been so difficult that skilled interpreters must have checked the satellite images manually because of the following issues. One is the requirement of the high detection accuracy rate. The other is the variety of the target, taking ships for example, there are many kinds of ships, such as boat, cruise ship, cargo ship, aircraft carrier, and so on. Furthermore, there are similar appearance objects throughout the image; therefore, it is often difficult even for the skilled interpreters to distinguish what object the pixels really compose. In this paper, we explore the feasibility of object extraction leveraging deep learning with high-resolution satellite images, especially focusing on ship detection. We calculated the detection accuracy using the WorldView-2 images. First, we collected the training images labelled as “ship” and “not ship”. After preparing the training data, we defined the deep neural network model to judge whether ships are existing or not, and trained them with about 50,000 training images for each label. Subsequently, we scanned the evaluation image with different resolution windows and extracted the “ship” images. Experimental result shows the effectiveness of the deep learning based object detection.

Dynamics modeling for sugar cane sucrose estimation using time series satellite imagery

Sugarcane, as one of the most mainstay crop in Brazil, plays an essential role in ethanol production. To monitor sugarcane crop growth and predict sugarcane sucrose content, remote sensing technology plays an essential role while accurate and timely crop growth information is significant, in particularly for large scale farming. We focused on the issues of sugarcane sucrose content estimation using time-series satellite image. Firstly, we calculated the spectral features and vegetation indices to make them be correspondence to the sucrose accumulation biological mechanism. Secondly, we improved the statistical regression model considering more other factors. The evaluation was performed and we got precision of 90% which is about 20% higher than the conventional method. The validation results showed that prediction accuracy using our sugarcane growth modeling and improved mix model is satisfied.

A spectral-spatial-dynamic hierarchical Bayesian (SSD-HB) model for estimating soybean yield

A method called a “spectral-spatial-dynamic hierarchical-Bayesian (SSD-HB) model,” which can deal with many parameters (such as spectral and weather information all together) by reducing the occurrence of multicollinearity, is proposed. Experiments conducted on soybean yields in Brazil fields with a RapidEye satellite image indicate that the proposed SSD-HB model can predict soybean yield with a higher degree of accuracy than other estimation methods commonly used in remote-sensing applications. In the case of the SSD-HB model, the mean absolute error between estimated yield of the target area and actual yield is 0.28 t/ha, compared to 0.34 t/ha when conventional PLS regression was applied, showing the potential effectiveness of the proposed model.

Automatic house detection from high-resolution satellite imagery

We have developed a house detection method based on machine learning for classification of houses and non-houses. In order to achieve precise classification, it is important to select features and to determine a dimensionality reduction method and a learning method. We first applied Gabor wavelet filters to generate the feature vectors and then developed a new method using the Adaboost algorithm to reduce the dimensionality of feature space. If a linear classifier made by one element of a feature vector is considered as a weak classifier in Adaboost, higher contribution dimensions can be selected. We used support vector machines (SVM) for the learning method. We evaluated our method by using QuickBird panchromatic images. Despite the significant variations in house shape and rooftop color, and in background clutter, our algorithm achieved high accuracy in house detection.