Research On Wheat Canopy Image Processing And Image Evaluation Index Of N Status

Rapidly and accurately monitoring crop nitrogen also gradually have became one of research hotspots in the field of precision agriculture development at home and abroad.In order to guide fertilization better through N status diagnosis at wheat jointing stage,the project would do some research about wheat canopy image processing and image evaluation index.In this study,it designed 1 meter and 90 degrees shooting plans in the field of natural environment,using the SLR camera to collect images of winter wheat canopy at different stages under different planting schemes from 2012 to 2014.Due to current domestic and foreign digital image evaluation index of different crops N status was relatively single and lack of targeted,the paper extracted color characteristic value of wheat leaf and analyzed of the correlation between image characteristics and LNC in the same period,and accordingly did research on the establishment method based on digital image processing technology and exploring new evaluation index;Although the shooting conditions artificially created under experimental field made sure that the sample of high quality,there still were several uneven illumination phenomena inevitably such as low contrast,facula and shadow,serious reflection under the natural light and complex background.Nevertheless,classic image segmentation algorithms existed phenomenon such as under-segmentation,over-segmentation.For the above problems,the main contributions of this paper are as follows:(1)NCMI was established as image evaluation index of N status diagnosis in wheat.The first step was to extract basic color characteristic value of canopy image samples in 2013 using the algorithm based on the normalized k-means clustering segmentation for the H component;the second step was to adjust and optimize the combination weights of the other remaining monochromatic components.A linear fitting was carried out which made use of basal image color characteristic values and leaf nitrogen concentration(LNC)measurement during the same period determined the best fitting content,the third step was to standardized and put forward normalized color mix index(NCMI).To study the feasibility of NCMI on monitoring nitrogen nutrition state,the experimental data in the year of 2014 were divided according to different periods and different cultivation schemes,and the correlation between NCMI and LNC,as well as that between other 3 typical image evaluation parameters,i.e.normalized red index(NRI),dark green color index(DGCI)and ratio of green light to red light(G/R)and LNC was analyzed quantitatively.The results showed that the correlation and the fitting of NCMI as winter wheat nitrogen nutrition evaluation index kept a good suitability,accuracy and stability,and meanwhile NCMI had a consistent change law with other 3 typical image characteristic indices(DGCI,G/R,NRI)under each scheme in 2014.Among them,the values of R2 between established NCMI and LNC during 3 sampling,higher than 3 contrast indicators at different degree,and the root mean square error(RMSE)respectively compared with the best RMSE of other 3 indices.Under the certain vegetation canopy coverage condition,the correlation between NCMI and LNC for the 2 cultivars(Shengxuan No.6 and Yangmai No.18)on March 31th and April 15th was better than that of the 3 comparative indices,and the lowest RMSE were 0.1833 and 0.2230,respectively;the related degree between NCMI and LNC under the planting density of D2(3.0×106 plant/hm2)in the 3 periods was higher,and the values of R2 and RMSE were superior to the 3 indices on March 8th,while they were consistent with those between DGCI and NRI at the other 2 stages;the RMSE between NCMI and LNC on April 15th was 0.1299,which was reduced by 5.18%compared with the lowest RMSE of the other 3 indicators;NCMI was also better than other indices under the N1 treatment(pure nitrogen 150 kg/hm2)of the D2 planting density,and the R2 increased by 7.69%and the RMSE improved by 4.11%compared with the best performer NRI.(2)An improved PCNN enhancement algorithm based on link intensity auto-setting(hereafter referred to as I_PCNN)and threshold setting and an image enhancement algorithm based on homomorphic filtering for the I component(hereafter referred to as I HomoFil),meanwhile they had a better effect reduction on the multiple mass of images caused by uneven illumination.In this article,degraded images accounted for 6.67%of samples was divided into two class.Category 1:low contrast of crown and soil,shadow under low light.Category 2:reflective,and light or shading under strong light.In view of the Category 1we used I PCNN to compensate light for the screen image of insufficient by light,weaken the influence of facula and shadow.two improvements were including a)link intensity valueβxy was adaptive setting,cut off the limitations of fixed value and workload;b)calculation method of dynamic threshold θxy was considering global and local characteristic pixels of space;Against problems of Category 2,we used I HomoFil designing a suitable filter to adjust the luminance component properly.The image enhanced was relatively uniform illumination,and it’s noise was reduced and kept more details of leaves,widening the gap between the target and background.(3)A k-means clustering segmentation algorithm based on image enhancement and angle model alpha was proposed(hereafter referred to as K-means_Imgenh&Angmodand)it had a better segmentation results of wheat under various drop quality and complex images by uneven illuminations.Firstly,to convert the wheat canopy image from RGB to HSI color space,and process the I component of HSI space by the adaptive enhancement algorithm,and properly adjust the saturation S component,so that it could compensate for the uneven distribution of the light intensity,remove the shadow and widen the contrast ratio.Secondly,to map processed enhanced image to the L*a*b*color space,extract a*,b*component,and establish alpha angle model which had smaller related degree with luminance component L*,was more suitable for being feature vector of k-means clustering.Finally,to process the enhanced wheat used k-means clustering segmentation.Segmentation experiment results of winter wheat canopy images which had differ light intensity,uneven illumination with before jointing stage and later jointing stage showed that:in allusion to wheat images of different illumination in different growth periods,the segmentation algorithm in this paper was better.To some degree,it could avoid over-segmentation phenomenon compared with the segmentation method of Otsu based on ExG;Compared with algorithm based on the normalized k-means clustering segmentation for the H weight,for the image of stronger reflectivity,it could also keep more complete details of wheat leaf and had better noise immunity of shadow,and for the image of the spot by the sun and the shadow of shade,it could segment more completely and accurately.(4)The accuracy of image segmentation was Quantitative analysised using image evaluation index of wheat nitrogen nutrition NCMI.By using K-means_Imgenh&Angmodand algorithm in this paper and k-means segmentation algorithm based on the H component in other paper,we respectively processed 12 low quality wheat canopy images and computed the NCMI.According to the correlation between NCMI and LNC we got,the R2 of NCMI established by this paper’s segmentation algorithm was 0.73,relatively higher 5.80%than in the R2 of NCMI established by k-means segmentation based on the H component and LNC,and the RMSE was 3.20,was relatively reduced by 5.60%.It quantitatively reflected that the algorithm in this paper was further reduced segmentation fault under the field environment through index of NCMI.The above-mentioned research results could provide the reference value for image evaluation index of winter wheat nitrogen nutrition diagnosis under certain canopy coverage;NCMI had important theoretical significance to subsequent research;K-means clustering segmentation algorithm based on image enhancement and model angle alpha could provide technical support for digital image processing techniques applied in the actual agricultural extension;It also provided a new way for related studies that we could quantitatively determine segmentation accuracy through image evaluation index.