基于自监督对比学习的寒旱区遥感图像河流识别方法

针对遥感图像河流数据样本人工标注成本高且难以大量获取,以及网络在河流图像边缘细节提取的效果不佳问题,提出一种通过自监督对比学习方式利用大量无标签遥感河流图像数据进行编码器预训练,并使用少量标签数据对预训练后的编码器进行微调,同时在编解码结构中使用一种新的非均匀采样方式的语义分割网络。自监督对比学习可以利用大量无标签数据进行前置任务模型训练,仅需少量标签数据对下游河流提取任务模型微调即可;非均匀采样方式能够通过对高频区域密集采样、对低频区域稀疏采样的方式获得图像中不同类别之间清晰的边界信息和同类别区域中的细节信息,减少模型的冗余度。在河流数据集上的实验表明,利用360幅有标签数据对预训练后的网络进行微调,其像素准确率、交并比、召回率分别达到90.4%、68.6%和83.2%,与使用1 200幅有标签数据训练的有监督AFR-LinkNet网络性能相当;在使用全部数据标签进行微调后,网络的像素准确率、交并比、召回率分别达到93.7%、73.2%和88.5%,相比AFR-LinkNet、DeepLabv3+、LinkNet、ResNet50和UNet网络,像素准确率分别提高3.1、7.6、12...     

All-around 3D plant modeling system using multiple images and its composition

In this study, we developed an all-around 3D plant modeling system that operates using images and is capable of measuring plants non-destructively without any contact. During the fabrication of this device, we selected a method capable of performing 3D model reconstruction from multiple images. We then developed an improved SfM-MVS (Structure from Motion / Multi-View-Stereo) method that enables 3D reconstruction by simply capturing images with a camera. The resulting image-based method offers a high degree of freedom because the hardware and software can comprise commercially available products, and it permits the use of one or more cameras according to the shape and size of the plant. The advantages of the image-based method are that 3D reconstruction can be conducted at any time as long as the images are already taken, and that the desired locations can be observed, measured, and analyzed from 2D images and a 3D point cloud. The device we developed is capable of 3D measurements and modeling of plants from a few millimeters to 2.4 m of height using this method. This article explains this device, the principles of its composition, and the accuracy of the models obtained from it.     

Detecting spruce and fir log diameter under bark by using shape and image data

The sawmilling industry stores and measures logs in bark in order to maximise efficiency, quality conservation and preservation. Billing is based on the diameter under bark, estimated based on manual or automatic bark detection. Although an automatic system is desirable for the industry, existing systems like tracheid effect scanners, X-ray or computed tomography either do not work reliably during all seasons or are very cost-intensive. This paper presents an approach for automatic determination of diameter under bark based on a multi-sensor approach including shape data, colour image data and tracheid effect data using laser scanning. For 686 spruce (Picea abies) and 79 fir (Abies alba) logs, diameters under bark are estimated and compared to the diameter after machine debarking. Estimation errors are close to zero and are below ±10 mm for more than 94% of the logs. Influences of season or characteristics like bark beetle are small. Specialised algorithms for diameter estimation in the presence of snow or bark beetle might further improve the result at the cost of required extra manual input.     

A Medical Image Quantization Method Suited for PACS

Image compression is very important in picture archiving and communication system(PACS). The author studied the statistical distribution of image wavelet subimage coefficients and concluded that the distribution of wavelet subimage coefficients is similar to that of Laplasian distribution. On the other hand, in image reconstruction, the coefficient with different amplitude owns different weight, and different accuracy can be applied to different coefficients according to their different weight. Then, the author has designed a image quantization encoding scheme for PACS. In this scheme, they selected the sample-standard-deviation of coefficients in every subimage as the quantization threshold, and accurately encoded those coefficients with higher weight. Also, this algorithm utilized the visual character of human. The test has proved that the main advantages of this method are the simplicity in computing and predictable encoded coefficients, and a high compression efficiency can obtain too.     

Parallel Optical Method in Laser Speckle Velocimetry and Particle Image Velocimetry

A method of parallel optical processing of double exposure specklegram or particle image using hololens array is presented.It can greatly decrease the processing time because the mechanical moving time of the X Y translation stage is decreased by tens of times.     

An Image Reconstruction Fast Algorithm Based on Orthogonal Wavelets Derived from Symmetric Scale Functions

A new image reconstruction algorithm is presented, based on orthogonal wavelets derived from symmetric scale functions. The analysis and practice show that the method lessens greatly the computational time,and improves the reconstruction quality.     

基于最小阈值法的棉花幼苗识别研究

 面对纯机械式分苗机构经常出现的漏分和多分的问题,本文提出使用基于最小阈值法的图像识别技术对棉花幼苗进行识别和分离。使用了HSV颜色空间内与光照条件无关的H值将RGB真彩色图像转化为灰度图像,以提高图像处理的实时性。基于最小值阈值法,提出一种能够有效地将处于复杂背景下的棉花幼苗茎部图像提取的方法,使用棉花幼苗茎部图像的面积和长宽比等几何信息能有效识别棉花幼苗,并得到棉花幼苗茎部的重心坐标。本文的研究证明了用图像处理中的最小阈值法识别棉花幼苗是可行的。     

用遥感影像提取大别山区水稻种植面积——以Landsat 8为例

本研究选取安徽省安庆市大别山为研究区,包括太湖县、岳西县、宿松县和潜山县,选取2016年和2017年的6景Landsat 8卫星遥感影像,通过遥感影像提取水稻种植面积,分析大别山区的水稻种植面积分布,并对研究区域的水稻种植面积进行动态实时监测。用遥感解译方法分别提取了研究区内的晚稻种植面积,并利用随机点验证和Kappa系数验证结果精度。结果表明:2017年和2016年水稻种植面积的提取精度分别为93.44%、93.78%,Kappa系数分别为0.86、0.83,证明水稻提取效果精确;对比研究区域各个县内的2017年和2016年晚稻种植面积,发现安庆市大别山区太湖县、潜山县、岳西县的水稻种植面积变化率均在5%以内,属于正常变化,由于2016年宿松县遭遇了水灾,部分农田被淹没,故宿松县2017年水稻种植面积相比2016年增加了13.01%;对比传统的农作物种植面积统计方法,利用遥感的方法更省人力、物力、财力,并且能精确、快速地实现对农作物的实时动态监测。     

基于GF-1卫星数据与面向对象分类的达里诺尔湿地自然保护区土地覆盖信息提取

以达里诺尔湿地自然保护区为研究区,基于国产GF-1遥感影像,采用面向对象和传统目视解译的分类方法对研究区土地覆盖遥感信息进行提取,并对其结果进行对比分析,采取混淆矩阵对面向对象分类结果进行精度验证。结果表明：（1）充分利用了GF-1遥感影像的光谱信息,面向对象分类采取试错法确定最优分割尺度为550,形状和紧致度因子分别为0.6和0.5,各波段权重均为1。（2）面向对象分类总体分类精度达98.22％,KAPPA系数为0.96;（3）面向对象分类方法可快速准确提取类型较为复杂的土地覆盖信息,为内陆湿地精准快速提取研究区土地覆盖分类信息提供参考,以期为湿地遥感业务化监测提供技术规范。     

基于多重分形去趋势移动平均分析的油菜叶片氮素营养诊断建模

为识别和诊断不同施氮水平下油菜叶片纹理特征,利用多重分形去趋势移动平均分析(MF-DMA),分别计算θ=0,θ=0.5,θ=1下油菜叶片图像的11个全局广义Hurst指数和其他6种相关的多重分形特征参数。利用不同的特征参数组合分别对基部叶片、中部叶片和顶部叶片进行氮素营养诊断识别。结果表明θ=0时诊断效果优于θ=0.5和θ=1;并且在θ=0时,基部叶片和中部叶片诊断效果优于顶部叶片,表明基部和中部叶片对氮素敏感性强于顶部。对3个部位混合的油菜样本进行施氮适中和亏缺两类定性诊断,结果表明支持向量机和随机森林这2种分类方法最佳,最佳识别准确率分别为95.81%和96.63%。说明该模型具有良好的有效性。