根据植物茎叶图像模拟根系图像的人工神经网络算法
|
摘要
以内蒙古野生甘草产区为试验地 ,采用图像关联的小波 -神经网络综合算法建立了甘草茎叶 -根系图像联结的 BP人工神经网络模型。该模型具有运算速度快 ,易于处理复杂图像数据的特点。通过植物茎叶图像对根系图像的模拟 ,实现块茎植物生物量数量化评估和农作物估产。
The patterns of botanic root system exist in an obvious correlation with the organs over the ground,such as leaves,branches,and stalks.Different species present varied features which mirror biological habits of a species being in correlation with the pattern of its own structure.Related researches have shown that root number,weight,and diameter offer a close relationship with the corresponding parts over the ground,so that it is possible to simulate the root system image with the leaf-stalk image.;In accordance with the principle of phytoecology,and based on biological habits of licorice,this paper recommended the wild licorice population structure in Chifeng region of the Inner Mongolia as a sample to establish a model of licorice artificial neuro-network linking up the leaves and stalks to its root system by using the technique of artificial neuro-nework integrated with image-information computer treatment.Through learning,training,and self-adaptation with the neuor-network,the ability to recognize and simulate the root system of the neuro-network got to be improved successively.It could randomize a single or multiple licorice leaf-stalk images to simulate the correspoding root system image by applying of the artificial neuro-network technique.A goal of simulating botanic root system image from vegetable leaf-stalk image was achieved at long.;On-the-spot inivestigation at the wild licorice growing area in Chifeng region,Inner Mongolia,the author selected both of 4 sampling units and premesured sample spots,of which the data of thd former were employed for the simulation and training of the neuro-network,and the latter for the assessment of distinguishing the artificial neuro-network and its imitation capacity.Around hundred of licorice leaf-stalk images and related data were gathered at the sample plots,meanwhile,the bio-variables related to some environmental parameters of ecological geography were collected too.;The chosen geographic environmental factors involved 4 categories:soil type,moisture state in growing time period,slope direction,and the elevation,after the factors to be digitized,they were inputted into the neuro-network model for computation.Digitized classification was grouped at 4 levels and the classified standards were determined to depend upon the concrete state of the trial sample plots.;The leaf-stalk and root system images of licorice are the major datum-source being employed for the artificial neuro-network model.The collection of patterns of the images were done standardizedly and formulatedly as well as possible.The photo-angle,height,and size were kept of showing no difference,and the color,light,and shade might be set strictlessly a little.;The simulated artifical neuro-network of licorice root system was designed to be a double-layer structure including each one of the input,hidden(mid-layer),and output layer.The transfer functions of the hidden and the output layer were the tansig and purelin type respectively.Thus,the output of data was realized in varied patterns.By auto-learning on counter propagation method and conducting the network to make a forward computation towards the sample input pattern,a comparison between the real output and the expected output of the network was made out,then adjusted the weighted average and the threshold value step by step until the faults got fulfilled to the requirement.To get pass of the above-mentioned understanding on the neuro-network and training well on the basis of weighted average,threshold value,and network structure,the simulation and prediction of a new sample can be approached successfully.;After getting the original image digitized,due to the data being very great and appearing of image noise,it couldn't be directly applied to learning and training for the neuro-network,for this reason,a wavelet image processing technique was recommended for the decomposition of licorice leaf-stalk image at 2~4 levels.The concrete way of doing was to divide the correlation leaf-stalk image into 4 groups,and each group was chosen of four typical images to rep
The patterns of botanic root system exist in an obvious correlation with the organs over the ground,such as leaves,branches,and stalks.Different species present varied features which mirror biological habits of a species being in correlation with the pattern of its own structure.Related researches have shown that root number,weight,and diameter offer a close relationship with the corresponding parts over the ground,so that it is possible to simulate the root system image with the leaf-stalk image.;In accordance with the principle of phytoecology,and based on biological habits of licorice,this paper recommended the wild licorice population structure in Chifeng region of the Inner Mongolia as a sample to establish a model of licorice artificial neuro-network linking up the leaves and stalks to its root system by using the technique of artificial neuro-nework integrated with image-information computer treatment.Through learning,training,and self-adaptation with the neuor-network,the ability to recognize and simulate the root system of the neuro-network got to be improved successively.It could randomize a single or multiple licorice leaf-stalk images to simulate the correspoding root system image by applying of the artificial neuro-network technique.A goal of simulating botanic root system image from vegetable leaf-stalk image was achieved at long.;On-the-spot inivestigation at the wild licorice growing area in Chifeng region,Inner Mongolia,the author selected both of 4 sampling units and premesured sample spots,of which the data of thd former were employed for the simulation and training of the neuro-network,and the latter for the assessment of distinguishing the artificial neuro-network and its imitation capacity.Around hundred of licorice leaf-stalk images and related data were gathered at the sample plots,meanwhile,the bio-variables related to some environmental parameters of ecological geography were collected too.;The chosen geographic environmental factors involved 4 categories:soil type,moisture state in growing time period,slope direction,and the elevation,after the factors to be digitized,they were inputted into the neuro-network model for computation.Digitized classification was grouped at 4 levels and the classified standards were determined to depend upon the concrete state of the trial sample plots.;The leaf-stalk and root system images of licorice are the major datum-source being employed for the artificial neuro-network model.The collection of patterns of the images were done standardizedly and formulatedly as well as possible.The photo-angle,height,and size were kept of showing no difference,and the color,light,and shade might be set strictlessly a little.;The simulated artifical neuro-network of licorice root system was designed to be a double-layer structure including each one of the input,hidden(mid-layer),and output layer.The transfer functions of the hidden and the output layer were the tansig and purelin type respectively.Thus,the output of data was realized in varied patterns.By auto-learning on counter propagation method and conducting the network to make a forward computation towards the sample input pattern,a comparison between the real output and the expected output of the network was made out,then adjusted the weighted average and the threshold value step by step until the faults got fulfilled to the requirement.To get pass of the above-mentioned understanding on the neuro-network and training well on the basis of weighted average,threshold value,and network structure,the simulation and prediction of a new sample can be approached successfully.;After getting the original image digitized,due to the data being very great and appearing of image noise,it couldn't be directly applied to learning and training for the neuro-network,for this reason,a wavelet image processing technique was recommended for the decomposition of licorice leaf-stalk image at 2~4 levels.The concrete way of doing was to divide the correlation leaf-stalk image into 4 groups,and each group was chosen of four typical images to rep
引文
[1] ZhangW H(张文辉),ZuY G(祖元刚).Study on the individual growth dynamic of adenophora lobophylla——Growth dynamicin one growing season,Bulletin ofBotanicalResearch( inChinese)(植物研究),1998,18(1):119～127.
[2] ZhangW H(张文辉),ZuY G(祖元刚).Study on the individual growth dynamic of adenophora lobophylla——Theroot system growth dynamic in the life span.Bulletin ofBotanicalResearch( inChinese)(植物研究),1998,18(1):128～135.
[3] HuangJH,etal.(黄建辉等).Study on bacteria population dynamics in cultural period and fish productivity inTuerjishanReservoirActaEcologicaSinica( inChinese)(生态学报),1999,19(2):270～277.
[4] ZhangR H,etal.(张仁华等).Studing ofCO2 flux on surface phants by remote sensing.Science inChina(SeriesD)( inChinese)(中国科学D辑),2000,30(2):215～224.
[5] JiangC X(蒋春曦).Distinguishing seismic wave and disturbing wave for local earthquake byBP neural net.Seis-mological andGeomagneticObservation andResearch( inChinese)(地震地磁观测与研究)1998,19(5):64～67.
[6] PiaoR M,etal.(朴荣民等).Application of theTheory ofTwo dimensionalWaveletTransformation toKillNoiseof theSeismicSignals.Journal ofShenyangInstitute ofChemicalTechnology( inChinese)(沈阳化工学院学报),1997,11(2):148～156.
[7] HongW,etal(洪伟,等).A study on the model of forestresources managementbased on the artificial neural net-work,Journal ofN aturalResources( inChinese)(自然资源学报),1998,13(1):69～72.
[8] ZuY G(祖元刚),ZaoZ H(赵则海),CongP T(丛沛桐),An elementary study on the distributional rule of canorygaps ofLarix gmelinii forests using waveletanalysis.ActaEcologicalSinica( inChinese)(生态学报),1999,19(6):927～931.
[9] ZuY G(祖元刚),ZaoZH(赵则海),CongP T(丛沛桐),Fractal analysis on spatial distribution patternsof popula-tion fromQuercusliaotunggensisforest inDanglingMountain,Bulletin ofBotanicalResearch( inChinese)(植物研究),2000:20(1):112～119.
[10] MichaelE.Hasselmo.Simulating nerves and networks.Science,1999,282(5391):1055～1058.
[11] sellersP J.etal.Modeling the exchanges of energy,water and carbon between continents and the atmosphere.Sci-ence,1999,275(5299):502～506.
[2] ZhangW H(张文辉),ZuY G(祖元刚).Study on the individual growth dynamic of adenophora lobophylla——Theroot system growth dynamic in the life span.Bulletin ofBotanicalResearch( inChinese)(植物研究),1998,18(1):128～135.
[3] HuangJH,etal.(黄建辉等).Study on bacteria population dynamics in cultural period and fish productivity inTuerjishanReservoirActaEcologicaSinica( inChinese)(生态学报),1999,19(2):270～277.
[4] ZhangR H,etal.(张仁华等).Studing ofCO2 flux on surface phants by remote sensing.Science inChina(SeriesD)( inChinese)(中国科学D辑),2000,30(2):215～224.
[5] JiangC X(蒋春曦).Distinguishing seismic wave and disturbing wave for local earthquake byBP neural net.Seis-mological andGeomagneticObservation andResearch( inChinese)(地震地磁观测与研究)1998,19(5):64～67.
[6] PiaoR M,etal.(朴荣民等).Application of theTheory ofTwo dimensionalWaveletTransformation toKillNoiseof theSeismicSignals.Journal ofShenyangInstitute ofChemicalTechnology( inChinese)(沈阳化工学院学报),1997,11(2):148～156.
[7] HongW,etal(洪伟,等).A study on the model of forestresources managementbased on the artificial neural net-work,Journal ofN aturalResources( inChinese)(自然资源学报),1998,13(1):69～72.
[8] ZuY G(祖元刚),ZaoZ H(赵则海),CongP T(丛沛桐),An elementary study on the distributional rule of canorygaps ofLarix gmelinii forests using waveletanalysis.ActaEcologicalSinica( inChinese)(生态学报),1999,19(6):927～931.
[9] ZuY G(祖元刚),ZaoZH(赵则海),CongP T(丛沛桐),Fractal analysis on spatial distribution patternsof popula-tion fromQuercusliaotunggensisforest inDanglingMountain,Bulletin ofBotanicalResearch( inChinese)(植物研究),2000:20(1):112～119.
[10] MichaelE.Hasselmo.Simulating nerves and networks.Science,1999,282(5391):1055～1058.
[11] sellersP J.etal.Modeling the exchanges of energy,water and carbon between continents and the atmosphere.Sci-ence,1999,275(5299):502～506.
版权所有：© 2023 中国地质图书馆 中国地质调查局地学文献中心