Time filter

Source Type

Fan X.,Beijing University of Technology | Fan X.,Cogent | Sun Y.,Beijing University of Technology | Yin B.,Beijing University of Technology | Guo X.,Chinese National Engineering Research Center for Information Technology in Agriculture
Pattern Recognition Letters | Year: 2010

Human fatigue is an important reason for many traffic accidents. To improve traffic safety, this paper proposes a novel Gabor-based dynamic representation for dynamics in facial image sequences to monitor human fatigue. Considering the multi-scale character of different facial behaviors, Gabor wavelets are employed to extract multi-scale and multi-orientation features for each image. Then features of the same scale are fused into a single feature according to two fusion rules to extract the local orientation information. To account for the temporal aspect of human fatigue, the fused image sequence is divided into dynamic units, and a histogram of each dynamic unit is computed and combined as dynamic features. Finally, AdaBoost algorithm is exploited to select the most discriminative features and construct a strong classifier to monitor fatigue. The proposed method was tested on a wide range of human subjects of different genders, poses and illuminations under real-life fatigue conditions. Experimental results show the validity of the proposed method, and an encouraging average correct rate is achieved. © 2009 Elsevier B.V. All rights reserved. Source

Chen Z.,University of Cambridge | Devereux B.,University of Cambridge | Gao B.,Chinese National Engineering Research Center for Information Technology in Agriculture | Amable G.,University of Cambridge
ISPRS Journal of Photogrammetry and Remote Sensing | Year: 2012

Airborne Lidar (Light detection and ranging) is an efficient tool for the generation of Digital Terrain Models (DTMs). Although many studies have been conducted in generating DTMs using Lidar data, it is still a challenging research area. The difficulty in filtering large buildings as well as a diversity of urban features makes the design of urban DTM generating methods an ongoing priority.This research adopted an upward-fusion methodology to generate urban DTMs using airborne Lidar data. Firstly, several preliminary DTMs of different resolutions were obtained using a local minimum method. Next, upward fusion was conducted between these DTMs. This process began with the DTM of the largest grid size, which was treated as a trend surface. A finer DTM was compared with this large scale DTM. By setting appropriate thresholds, a new DTM was achieved by selecting qualified elevation values from the finer DTM and retaining the value of the trend surface when the value from the finer DTM was beyond the threshold. This process continued iteratively until all preliminary DTMs had been included in the upward fusion and a refined DTM of high resolution was achieved. To further reduce possible errors in the resulting DTM, an extended local minimum method was proposed for filtering outliers and generating preliminary DTMs.A case study was carried out in the city of Cambridge, which represents an urban landscape with a variety of building forms, street patterns and vegetation structures. The time efficiency, results of the accuracy assessments and comparison with leading software proved the success of the case study and indicated that upward-fusion was an effective method for the generation of urban DTMs with airborne Lidar data and could improve the accuracy of other DTM generating algorithms. This paper also proposed possible approaches for further improvements on this methodology. © 2012 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Source

Yin X.-C.,University of Science and Technology Beijing | Huang K.,Xian Jiaotong - Liverpool University | Hao H.-W.,CAS Institute of Automation | Iqbal K.,University of Science and Technology Beijing | Wang Z.-B.,Chinese National Engineering Research Center for Information Technology in Agriculture
Neurocomputing | Year: 2014

We consider the classifier ensemble problem in this paper. Due to its superior performance to individual classifiers, class ensemble has been intensively studied in the literature. Generally speaking, there are two prevalent research directions on this, i.e., to diversely generate classifier components, and to sparsely combine multiple classifiers. While most current approaches are emphasized on either sparsity or diversity only, we investigate the classifier ensemble by learning both sparsity and diversity simultaneously. We manage to formulate the classifier ensemble problem with the sparsity or/and diversity learning in a general framework. In particular, the classifier ensemble with sparsity and diversity can be represented as a mathematical optimization problem. We then propose a heuristic algorithm, capable of obtaining ensemble classifiers with consideration of both sparsity and diversity. We exploit the genetic algorithm, and optimize sparsity and diversity for classifier selection and combination heuristically and iteratively. As one major contribution, we introduce the concept of the diversity contribution ability so as to select proper classifier components and evolve classifier weights eventually. Finally, we compare our proposed novel method with other conventional classifier ensemble methods such as Bagging, least squares combination, sparsity learning, and AdaBoost, extensively on UCI benchmark data sets and the Pascal Large Scale Learning Challenge 2008 webspam data. The experimental results confirm that our approach leads to better performance in many aspects. © 2014 Elsevier B.V. Source

Li B.,Chinese National Engineering Research Center for Information Technology in Agriculture | Wang M.,China Agricultural University
Intelligent Automation and Soft Computing | Year: 2013

Fruit recognition and navigation path extraction are important issues for developing fruit harvesting robots. This manuscript presents a recent study on developing an algorithm for recognizing "on-the-go" pineapple fruits and the cultivation rows for a harvesting robotic system. In-field pineapple recognition can be difficult due to many overlapping leaves from neighbouring plants. As pineapple fruits (Ananas comosus) are normally located at top of the plant with a crowned by a compact tuft of young leaves, image processing algorithms were developed to recognize the crown to locate the corresponding pineapple fruit in this study. RGB (Red, Green, and Blue) images were firstly collected from top-view of pineapple trees in the field and transformed into HSI (Hue, Saturation and Intensity) colour model. Then, Features of pineapple crowns were extracted and used for developing a classification algorithm. After the pineapple crowns were recognized, locations of the crowns grown in one row were determined and linearly fitted into a line, which could be used for navigating the harvesting robots to conduct the harvest. To validate the above algorithms, 100 images were taken in a pineapple field under different environments in Guangdong province as a validation set. The results showed that pineapple recognition rate can reach 94% on clear sky day, which was much better than that on overcast sky day and the navigation path was well fitted. © 2013 Copyright TSI® Press. Source

Yang G.,Chinese National Engineering Research Center for Information Technology in Agriculture | Zhao C.,Beijing Normal University | Liu Q.,Beijing Normal University | Huang W.,Beijing Normal University | Wang J.,Beijing Normal University
IEEE Transactions on Geoscience and Remote Sensing | Year: 2011

This paper presents a new forest leaf area index (LAI) inversion method from multisource and multiangle data combined with a radiative transfer model and the strategy of k-means clustering and artificial neural network (ANN). Four scenes of Landsat-5 Thematic Mapper (L5TM) and Beijing-1 small satellite multispectral sensors (BJ1) images, acquired at different times, were selected to construct multisource and multiangle image data in this study. Considering a vertical distribution of forest LAI from both overstory and understory, a hybrid model of the invertible forest reflectance model (INFORM) was used to support the retrieval of forest LAI to eliminate the dependence of understory vegetation. The simulated data from INFORM outputs, added with a random noise, were first clustered by k-means method, and were then trained by ANN to obtain the inversion model for each group (cluster). Next, the inversion model was applied to the different combinations of multiangle data to retrieve the forest LAI. Finally, a validation of inverted results with Moderate Resolution Imaging Spectroradiometer LAI product and field measurements was conducted. The experimental results indicate that the accuracy of the inverted forest LAI can be improved through the addition of observation angle data, if the quality of the image data is ensured. The inversion accuracy of LAI with the multiangle image data is improved by 30% compared to the average accuracy of the inverted LAI with the single angle data after considering the addition of random noise to the ANN training data. © 2006 IEEE. Source

Discover hidden collaborations