Void ratio measures compactness of ground soil in geotechnical engineering. When samples are collected in certain area for mapping void ratios, other relevant types of properties such as water content may be also anal...Void ratio measures compactness of ground soil in geotechnical engineering. When samples are collected in certain area for mapping void ratios, other relevant types of properties such as water content may be also analyzed. To map the spatial distribution of void ratio in the area based on these types of point, observation data interpolation is often needed. Owing to the variance of sampling density along the horizontal and vertical directions, special consideration is required to handle anisotropy of estimator. 3D property modeling aims at predicting the overall distribution of property values from limited samples, and geostatistical method can be employed naturally here because they help to minimize the mean square error of estimation. To construct 3D property model of void ratio, cokriging was used considering its mutual correlation with water content, which is another important soil parameter. Moreover, K-D tree was adopted to organize the samples to accelerate neighbor query in 3D space during the above modeling process. At last, spatial configuration of void ratio distribution in an engineering body was modeled through 3D visualization, which provides important information for civil engineering purpose.展开更多
This paper describes three algorithms for retrieving precipitation over oceans from brightness temperatures (TBs) of the Micro-Wave Humidity Sounder-2 (MHWS-2) onboard Fengyun-3C (FY-3C). For algorithm development, sc...This paper describes three algorithms for retrieving precipitation over oceans from brightness temperatures (TBs) of the Micro-Wave Humidity Sounder-2 (MHWS-2) onboard Fengyun-3C (FY-3C). For algorithm development, scattering- induced TB depressions (ΔTBs) of MWHS-2 at channels between 89 and 190 GHz were collocated to rain rates derived from measurements of the Global Precipitation Measurement’s Dual-frequency Precipitation Radar (DPR) for the year 2017. ΔTBs were calculated by subtracting simulated cloud-free TBs from bias-corrected observed TBs for each channel. These ΔTBs were then related to rain rates from DPR using (1) multilinear regression (MLR);the other two algorithms, (2) range searches (RS) and (3) nearest neighbor searches (NNS), are based on k-dimensional trees. While all three algorithms produce instantaneous rain rates, the RS algorithm also provides the probability of precipitation and can be understood in a Bayesian framework. Different combinations of MWHS-2 channels were evaluated using MLR and results suggest that adding 118 GHz improves retrieval performance. The optimal combination of channels excludes high-peaking channels but includes 118 GHz channels peaking in the mid and high troposphere. MWHS-2 observations from another year were used for validation purposes. The annual mean 2.5° × 2.5° gridded rain rates from the three algorithms are consistent with those from the Global Precipitation Climatology Project (GPCP) and DPR. Their correlation coefficients with GPCP are 0.96 and their biases are less than 5%. The correlation coefficients with DPR are slightly lower and the maximum bias is ~8%, partly due to the lower sampling density of DPR compared to that of MWHS-2.展开更多
A literature analysis has shown that object search,recognition,and tracking systems are becoming increasingly popular.However,such systems do not achieve high practical results in analyzing small moving living objects...A literature analysis has shown that object search,recognition,and tracking systems are becoming increasingly popular.However,such systems do not achieve high practical results in analyzing small moving living objects ranging from 8 to 14 mm.This article examines methods and tools for recognizing and tracking the class of small moving objects,such as ants.To fulfill those aims,a customized You Only Look Once Ants Recognition(YOLO_AR)Convolutional Neural Network(CNN)has been trained to recognize Messor Structor ants in the laboratory using the LabelImg object marker tool.The proposed model is an extension of the You Only Look Once v4(Yolov4)512×512 model with an additional Self Regularized Non–Monotonic(Mish)activation function.Additionally,the scalable solution for continuous object recognizing and tracking was implemented.This solution is based on the OpenDatacam system,with extended Object Tracking modules that allow for tracking and counting objects that have crossed the custom boundary line.During the study,the methods of the alignment algorithm for finding the trajectory of moving objects were modified.I discovered that the Hungarian algorithm showed better results in tracking small objects than the K–D dimensional tree(k-d tree)matching algorithm used in OpenDataCam.Remarkably,such an algorithm showed better results with the implemented YOLO_AR model due to the lack of False Positives(FP).Therefore,I provided a new tracker module with a Hungarian matching algorithm verified on the Multiple Object Tracking(MOT)benchmark.Furthermore,additional customization parameters for object recognition and tracking results parsing and filtering were added,like boundary angle threshold(BAT)and past frames trajectory prediction(PFTP).Experimental tests confirmed the results of the study on a mobile device.During the experiment,parameters such as the quality of recognition and tracking of moving objects,the PFTP and BAT,and the configuration parameters of the neural network and boundary line model were analyzed.The results showed an increased tracking accuracy with the proposed methods by 50%.The study results confirmed the relevance of the topic and the effectiveness of the implemented methods and tools.展开更多
This paper proposes an efficient retrieval ap- proach for iris using local features. The features are extracted from segmented iris image using scale invariant feature trans- form （SIFT）. The keypoint descriptors ex...This paper proposes an efficient retrieval ap- proach for iris using local features. The features are extracted from segmented iris image using scale invariant feature trans- form （SIFT）. The keypoint descriptors extracted from SIFT are clustered into m groups using k-means. The idea is to perform indexing of keypoints based on descriptor property. During database indexing phase, k-d tree k-dimensional tree is constructed for each cluster center taken from N iris im- ages. Thus for m clusters, rn such k-d trees are generated de- noted as ti, where 1 ~〈 i ~〈 m. During the retrieval phase, the keypoint descriptors from probe iris image are clustered into m groups and ith cluster center is used to traverse correspond- ing ti for searching, k nearest neighbor approach is used, which finds p neighbors from each tree （ti） that falls within certain radius r centered on the probe point in k-dimensional space. Finally, p neighbors from m trees are combined using union operation and top S matches （S c_ （m x p）） correspond- ing to query iris image are retrieved. The proposed approach has been tested on publicly available databases and outper- forms the existing approaches in terms of speed and accuracy.展开更多
基金supported by Beijing Multi-parameters 3D Geological Survey Program (No. 200313000045)
文摘Void ratio measures compactness of ground soil in geotechnical engineering. When samples are collected in certain area for mapping void ratios, other relevant types of properties such as water content may be also analyzed. To map the spatial distribution of void ratio in the area based on these types of point, observation data interpolation is often needed. Owing to the variance of sampling density along the horizontal and vertical directions, special consideration is required to handle anisotropy of estimator. 3D property modeling aims at predicting the overall distribution of property values from limited samples, and geostatistical method can be employed naturally here because they help to minimize the mean square error of estimation. To construct 3D property model of void ratio, cokriging was used considering its mutual correlation with water content, which is another important soil parameter. Moreover, K-D tree was adopted to organize the samples to accelerate neighbor query in 3D space during the above modeling process. At last, spatial configuration of void ratio distribution in an engineering body was modeled through 3D visualization, which provides important information for civil engineering purpose.
基金This work was supported by a NASA grant(Grant No.NNX17AJ09G)to Vanderbilt University.
文摘This paper describes three algorithms for retrieving precipitation over oceans from brightness temperatures (TBs) of the Micro-Wave Humidity Sounder-2 (MHWS-2) onboard Fengyun-3C (FY-3C). For algorithm development, scattering- induced TB depressions (ΔTBs) of MWHS-2 at channels between 89 and 190 GHz were collocated to rain rates derived from measurements of the Global Precipitation Measurement’s Dual-frequency Precipitation Radar (DPR) for the year 2017. ΔTBs were calculated by subtracting simulated cloud-free TBs from bias-corrected observed TBs for each channel. These ΔTBs were then related to rain rates from DPR using (1) multilinear regression (MLR);the other two algorithms, (2) range searches (RS) and (3) nearest neighbor searches (NNS), are based on k-dimensional trees. While all three algorithms produce instantaneous rain rates, the RS algorithm also provides the probability of precipitation and can be understood in a Bayesian framework. Different combinations of MWHS-2 channels were evaluated using MLR and results suggest that adding 118 GHz improves retrieval performance. The optimal combination of channels excludes high-peaking channels but includes 118 GHz channels peaking in the mid and high troposphere. MWHS-2 observations from another year were used for validation purposes. The annual mean 2.5° × 2.5° gridded rain rates from the three algorithms are consistent with those from the Global Precipitation Climatology Project (GPCP) and DPR. Their correlation coefficients with GPCP are 0.96 and their biases are less than 5%. The correlation coefficients with DPR are slightly lower and the maximum bias is ~8%, partly due to the lower sampling density of DPR compared to that of MWHS-2.
文摘A literature analysis has shown that object search,recognition,and tracking systems are becoming increasingly popular.However,such systems do not achieve high practical results in analyzing small moving living objects ranging from 8 to 14 mm.This article examines methods and tools for recognizing and tracking the class of small moving objects,such as ants.To fulfill those aims,a customized You Only Look Once Ants Recognition(YOLO_AR)Convolutional Neural Network(CNN)has been trained to recognize Messor Structor ants in the laboratory using the LabelImg object marker tool.The proposed model is an extension of the You Only Look Once v4(Yolov4)512×512 model with an additional Self Regularized Non–Monotonic(Mish)activation function.Additionally,the scalable solution for continuous object recognizing and tracking was implemented.This solution is based on the OpenDatacam system,with extended Object Tracking modules that allow for tracking and counting objects that have crossed the custom boundary line.During the study,the methods of the alignment algorithm for finding the trajectory of moving objects were modified.I discovered that the Hungarian algorithm showed better results in tracking small objects than the K–D dimensional tree(k-d tree)matching algorithm used in OpenDataCam.Remarkably,such an algorithm showed better results with the implemented YOLO_AR model due to the lack of False Positives(FP).Therefore,I provided a new tracker module with a Hungarian matching algorithm verified on the Multiple Object Tracking(MOT)benchmark.Furthermore,additional customization parameters for object recognition and tracking results parsing and filtering were added,like boundary angle threshold(BAT)and past frames trajectory prediction(PFTP).Experimental tests confirmed the results of the study on a mobile device.During the experiment,parameters such as the quality of recognition and tracking of moving objects,the PFTP and BAT,and the configuration parameters of the neural network and boundary line model were analyzed.The results showed an increased tracking accuracy with the proposed methods by 50%.The study results confirmed the relevance of the topic and the effectiveness of the implemented methods and tools.
文摘This paper proposes an efficient retrieval ap- proach for iris using local features. The features are extracted from segmented iris image using scale invariant feature trans- form （SIFT）. The keypoint descriptors extracted from SIFT are clustered into m groups using k-means. The idea is to perform indexing of keypoints based on descriptor property. During database indexing phase, k-d tree k-dimensional tree is constructed for each cluster center taken from N iris im- ages. Thus for m clusters, rn such k-d trees are generated de- noted as ti, where 1 ~〈 i ~〈 m. During the retrieval phase, the keypoint descriptors from probe iris image are clustered into m groups and ith cluster center is used to traverse correspond- ing ti for searching, k nearest neighbor approach is used, which finds p neighbors from each tree （ti） that falls within certain radius r centered on the probe point in k-dimensional space. Finally, p neighbors from m trees are combined using union operation and top S matches （S c_ （m x p）） correspond- ing to query iris image are retrieved. The proposed approach has been tested on publicly available databases and outper- forms the existing approaches in terms of speed and accuracy.
