Jiansheng Chen

Remote Sensing Scene Classification Based on Convolutional Neural Networks Pre-Trained Using Attention-Guided Sparse Filters

Semantic-level land-use scene classification is a challenging problem, in which deep learning methods, e.g., convolutional neural networks (CNNs), have shown remarkable capacity. However, a lack of sufficient labeled images has proved a hindrance to increasing the land-use scene classification accuracy of CNNs. Aiming at this problem, this paper proposes a CNN pre-training method under the guidance of a human visual attention mechanism. Specifically, a computational visual attention model is used to automatically extract salient regions in unlabeled images. Then, sparse filters are adopted to learn features from these salient regions, with the learnt parameters used to initialize the convolutional layers of the CNN. Finally, the CNN is further fine-tuned on labeled images. Experiments are performed on the UCMerced and AID datasets, which show that when combined with a demonstrative CNN, our method can achieve 2.24% higher accuracy than a plain CNN and can obtain an overall accuracy of 92.43% when combined with AlexNet. The results indicate that the proposed method can effectively improve CNN performance using easy-to-access unlabeled images and thus will enhance the performance of land-use scene classification especially when a large-scale labeled dataset is unavailable.

A Spatio-Temporal Model for Forest Fire Detection Using HJ-IRS Satellite Data

Fire detection based on multi-temporal remote sensing data is an active research field. However, multi-temporal detection processes are usually complicated because of the spatial and temporal variability of remote sensing imagery. This paper presents a spatio-temporal model (STM) based forest fire detection method that uses multiple images of the inspected scene. In STM, the strong correlation between an inspected pixel and its neighboring pixels is considered, which can mitigate adverse impacts of spatial heterogeneity on background intensity predictions. The integration of spatial contextual information and temporal information makes it a more robust model for anomaly detection. The proposed algorithm was applied to a forest fire in 2009 in the Yinanhe forest, Heilongjiang province, China, using two-month HJ-1B infrared camera sensor (IRS) images. A comparison of detection results demonstrate that the proposed algorithm described in this paper are useful to represent the spatio-temporal information contained in multi-temporal remotely sensed data, and the STM detection method can be used to obtain a higher detection accuracy than the optimized contextual algorithm.

A Method of Obtaining Accurate Active Area of Remote Sensing Image and Application in Mosaicking

The generation of active area especially the irregular active area can always happen during the geometry of affine transformation in image geometric correction and clipping along the administrative area shape file, but there are few methods or algorithms in order to obtain the active area. The article analyses the current situation in extracting the active area, designs an algorithm of obtaining irregular active area based on Convex Hull Algorithms and implements a Mosaic module for remote sensing image based on the irregular active area extracting algorithm. The experiment proves that the effect of the articles mosaic method is better than the main current commerce remote sensing image processing software, and with the increase of percentage of background area, the articles mosaic method can reduce the processing time.

Long Short-Term Memory Neural Networks for Online Disturbance Detection in Satellite Image Time Series

A satellite image time series (SITS) contains a significant amount of temporal information. By analysing this type of data, the pattern of the changes in the object of concern can be explored. The natural change in the Earth’s surface is relatively slow and exhibits a pronounced pattern. Some natural events (for example, fires, floods, plant diseases, and insect pests) and human activities (for example, deforestation and urbanisation) will disturb this pattern and cause a relatively profound change on the Earth’s surface. These events are usually referred to as disturbances. However, disturbances in ecosystems are not easy to detect from SITS data, because SITS contain combined information on disturbances, phenological variations and noise in remote sensing data. In this paper, a novel framework is proposed for online disturbance detection from SITS. The framework is based on long short-term memory (LSTM) networks. First, LSTM networks are trained by historical SITS. The trained LSTM networks are then used to predict new time series data. Last, the predicted data are compared with real data, and the noticeable deviations reveal disturbances. Experimental results using 16-day compositions of the moderate resolution imaging spectroradiometer (MOD13Q1) illustrate the effectiveness and stability of the proposed approach for online disturbance detection.

Visible and infrared image registration algorithm based on NSCT and gradient mirroring

Multi-sensor image registration is an important part of the remote sensing image processing. The gray property of the same object would have large differences in infrared and visible imaging mode, so it could get less matching points by using traditional SIFT algorithm directly in registration. However, NSCT decomposition can represent the structural information of the image very well and extract more SIFT feature points in its high frequency decomposed image. In addition, traditional SIFT descriptors’ gradient is affected by gray contrast, which could get less feature matching points during the similarity search in the matching procedure. Gradient mirroring (GM) is a method that can modify the direction of the feature points, which can reduce the contrast impact on the similarity matching. Therefore, a novel method combining NSCT and GM is proposed in this article. The experiments prove that, comparing with the traditional SIFT algorithm, the new method can get more matching points, better distributing and higher matching rate in infrared and visible image registration.

Fast detection of changed blocks in land use map

Using remote sensing images to monitor land use information, extract changed information, which has a wide range of research value and is the key technologies in the field of land use monitoring. The article analyses the current methods of change detection and designs a fast detection method of changed blocks in land use map with the before and after remote sensing images. The method uses the spirit of object-oriented technology, and makes the polygons in the land use map directly as the objects for research. Then calculates the feature value(mean value of layer 1 and layer 2, NDVI, GLCM Homogeneity of layer 3 and layer 4 used in this article ) of each objects in the before and after remote sensing images. After normalized processing for feature value, calculates the Euclidean distance between feature vectors. Finally, extracts the changed polygons in land use map according to the threshold. The experiment proves that the articles method can fast and accurately detect the changed blocks in land use map.

Decision tree coupled with feature optimization for object-based classification of ZY-1-02C satellite images

The Separability and Thresholds (SEaTH) algorithm calculates the the SEparability and the corresponding THresholds of object classes for any number of given features. However, it is applicable only to the normally distributed training data. To cope with the problem, The Classification And Regression Tree (CART) coupled with SEaTH for object-based classification approach is proposed in the paper. The idea of this method is derived from the merits of the CART which can effectively analyze the non-normally distributed data and automatically create the classification tree. A comparison of classification results demonstrate that the solution for object-based classification proposed in this article can be used to obtain a higher classification accuracy than SEaTH classification.

Practical Bottom-up Golf Course Detection Using Multispectral Remote Sensing Imagery

The rapid growth of golf course has constituted a nonnegligible threat to conservation of cropland and water resource in China. To monitor golf course at a large scale with low cost, a practical bottom-up golf course detection approach using multispectral remote sensing imagery is proposed. First of all, turfgrass, water-body and bunker are determined as the basic elements based on analyzing golf course land-use characteristics. Secondly, turfgrass and water-body are extracted using spectral indexes and these two basic elements are combined as region-of-interest under guidance of prior-knowledge. Afterwards, bunker is extracted by spectral mixture analysis restricted to region-of-interest. Finally, fuzzy C-means is adopted to recognize golf course using landscape metrics. A SPOT-5 HRG multispectral image of Beijing is used to validate the proposed method, and detection rate and false alarm rate are 86.67% and 38.10% respectively.

Knowledge-guided golf course detection using a convolutional neural network fine-tuned on temporally augmented data

Abstract. The tremendous success of deep learning models such as convolutional neural networks (CNNs) in computer vision provides a method for similar problems in the field of remote sensing. Although research on repurposing pretrained CNN to remote sensing tasks is emerging, the scarcity of labeled samples and the complexity of remote sensing imagery still pose challenges. We developed a knowledge-guided golf course detection approach using a CNN fine-tuned on temporally augmented data. The proposed approach is a combination of knowledge-driven region proposal, data-driven detection based on CNN, and knowledge-driven postprocessing. To confront data complexity, knowledge-derived cooccurrence, composition, and area-based rules are applied sequentially to propose candidate golf regions. To confront sample scarcity, we employed data augmentation in the temporal domain, which extracts samples from multitemporal images. The augmented samples were then used to fine-tune a pretrained CNN for golf detection. Finally, commission error was further suppressed by postprocessing. Experiments conducted on GF-1 imagery prove the effectiveness of the proposed approach.

Vehicles detection using GF-2 imagery based on watershed image segmentation

Road traffic volume monitoring plays an important role in transportation planning and spatial development, particularly in urban areas. The high-resolution satellite imagery provides a new data source to detect vehicles. Meanwhile, Satellite image covers large areas instantaneously, providing a possibility for snapshotting road traffic conditions. In this paper, we proposed an approach based on watershed image segmentation to detect the urban road vehicles from GF-2 imagery. The vehicles detection involves the two main steps: Firstly, a GIS road vector map and vegetation masks were applied to the image to guide vehicle detection by restricting the roads only. Secondly, watershed image segmentation was performed to separate bright and dark vehicles from the background in the road region. Then, a rule-based classifier was established to classify the image objects into the vehicle and the non-vehicle objects by using the spectral and shape feature information of image objects. Finally, the overall performance of the vehicle detection were compared with the manually counts, yielding overall accuracy of 81% with 93% classification accuracy. This detection accuracy may be considered acceptable for operational use in traffic monitoring.