| Desert steppe,an important part of grassland in Inner Mongolia,has been degrading seriously in recent years,due to the influence of natural or human activities and other factors.This has caused serious impacts on the sustainable development of the local ecological environment.It is particularly important to analyze and predict the potential degradation risk of desert steppe,and then to strengthen the protection and management of Desert steppe.With the rapid development of digital image processing technology,it has been gradually used in the research of grassland ecological environment and grassland vegetation.The field experiment was conducted in a desert steppe in Ordos and the indicative plant named Cleistogenes Songorica in this area was chosen as the experimental object.In this dissertation,the Cleistogenes Songorica leaf anatomical structure was analyzed by image processing techniques.The key technologies such as image enhancement,image segmentation,feature extraction and parameter measurement in the microscopic image processing were studied emphatically.The experiment revealed the response and adaptation mechanism of Cleistogenes Songorica leaf anatomical structure to the desert steppe,and also studied the correlation between the structural indicators and the degradation gradient of grassland.The main research work is as follows:(1)Image acquisitionBased on the popular classification theory proposed by Academician Li Bo for grassland degradation,the experimental plots with different degradation gradients were determined by field and group research.To meet the requirements of statistics and ecology,a total of 10800 samples were collected in in July to August of the years from 2014 to 2016,and 768 sections were obtained by paraffin section.Then the images of the Cleistogenes Songorica leaf anatomical structure were acquired and stored by the YYS-80E type biological optical microscope and computer.(2)Image enhancement.To avoid the poor contrast enhancement performance of CLAHE algorithm on over-dark images,a contrast enhancement algorithm based on adaptive brightness adjustment was proposed to introduce the global brightness adaptive adjustment of image on the basis of the CLAHE algorithm.Compared with the original images,the values of image information entropy,image average gradient and image contrast of 30 test images after enhanced by this algorithm were improved about 22.3%,177.7%and 67%respectively,while these were improved about 18.4%,164.4%and 59.6%respectively by CLAHE algorithm.(3)Image segmentation.?The integral segmentation.The contrast and details of the Cleistogenes Songorica leaf anatomical structure image were decreased obviously after disposed by the traditional grayscale methods.To solve this problem,a grayscale method was proposed for the Cleistogenes Songorica leaf anatomical structure image by adjusting the weight of R,G and B,according to the color difference of red and blue between the foreground and the background.A comparison was made between the new grayscale method and the traditional ones.It showed that compared with the component method(R),which performed better in the traditional methods,the values of image information entropy,image average gradient and image contrast were improved about 1.05%,224.4%and 133.7%respectively by using this method.On the basis of this,the integral segmentation result of the Cleistogenes Songorica leaf anatomical structure was obtained by using the maximum inter-class variance(Otsu)combined with morphological open operation and linear filtering.In 30 test images,the average false positive rate was 0.75%,the average false negative rate was 3.49%and the average global segmentation accuracy was 98.14%.?The cuticle segmentation.In view of the fact that the cuticle of the Cleistogenes Songorica leaf anatomical structure was dyed red and it had strongest visual saliency in the whole image,this dissertation proposed a saliency detection algorithm combining AC and FT to realize the accurate segmentation of the cuticle of the Cleistogenes Songorica leaf anatomical structure,based on deeply studied the visual saliency rules,the image characteristic and the processing mode of the ten classical visual saliency detection algorithms.In the 30 test images segmented after visual saliency detection respectively by FT algorithm,AC algorithm and this algorithm,the average false positive rate FPR were 0.75%,1.02%and 0.48%,the average false negative rate FNR were 7.19%,5.87%and 3.45%,and the average global segmentation accuracy GSA were 98.43%,98.11%and 99.21%.Thus it can be seen that the segmentation accuracy of this algorithm was higher that of the other two algorithms.?The vascular bundle segmentation.Owing to the fact that the overall average color value of vascular bundle sheath cells was the highest in the whole image,this dissertation proposed an adaptive threshold segmentation algorithm in RGB spatial based on automatic selection of sample points.This algorithm automatically selected the pixel value of the location of vascular bundle sheath cells as the color sample points applying the image arithmetic operation and the threshold segmentation.At the same time,the accurate segmentation of vascular bundle was realized by measuring the similarity by Euclidean distance between the average color estimation value of the sample points and the pixel value in the image to be segmented.In the 30 test images segmented respectively by the traditional algorithm and the proposed one,the average false positive rate FPR were 7.61%and 0.72%,the average false negative rate FNR were 7.08%and 1.35%,and the average global segmentation accuracy GSA were 92.64%and 98.73%.Thus it can be seen that the segmentation accuracy of this algorithm was higher than that of the traditional one.?The kranz mesophyll cell segmentation.According to the image characteristics of kranz mesophyll cell image of the Cleistogenes Songorica leaf anatomical structure,this dissertation proposed to realize the kranz mesophyll cell segmentation by an interactive image segmentation method.The active contour algorithm based on the variational level set and the interactive image segmentation by maximal similarity based region merging,the two best interactive image segmentation algorithms at present,were compared and analyzed.The results of 30 test images segmented respectively by the two algorithms respectively showed that the average false positive rate FPR were 1.63%and 0.82%,the average false negative rate FNR were 4.01%and 4%,and the average global segmentation accuracy GSA were 98.20%and 98.65%.The segmentation accuracy of the interactive image segmentation by maximal similarity based region merging was slightly higher than that of the active contour algorithm based on the variational level set.Besides,in terms of the operation and the stability of the algorithm,it was more suitable for the kranz mesophyll cell segmentation.(4)Feature extraction?The leaf width(LW)measurement.The actual measurement mode of LW was to measure the longest distance between the two points on the Cleistogenes Songorica leaf anatomical structure image.The convex hull was constructed by Graham algorithm and the minimum external rectangle of the integral leaf anatomical structure image was obtained,thus,the value of leaf width was achieved.The measurement was compared with its mean value of interactive measurement,the average relative error of 30 test images was 0.96%and the average time consumed was 4.87s by the measurement proposed in this dissertation.?The leaf thickness(TL)measurement.The actual measurement of TL was to measure the distance between the concave points and the concave points on the left and right boundary of the Cleistogenes Songorica leaf anatomical structure image,it was also the distance between the convex points and the convex points.This dissertation obtained the polygon outline and apexes of the whole leaf anatomical structure image by using the minimum perimeter polygons approximate method,then to detect the concave and convex points.While there were many useless points detected by the concave and convex points detection algorithm based on vector product method.To avoid this problem,this dissertation proposed concave and convex point detection algorithm combined the corner detection algorithm and vector product method to eliminate the useless apexes to obtain the concave and convex points in accordance with the actual measurement.Then,the leaf thickness value(TL)was obtained by concave points matching and convex points matching.The measurement was compared with its mean value of interactive measurement,the average relative error of 30 test images was 3.69%and the average time consumed was 4.92s by the measurement proposed in this dissertation.?The measurement of the thickness of kranz mesophyll cell(TKMC)and the diameter of vascular bundle(TVB).The shape of kranz mesophyll cell and the vascular bundle of the Cleistogenes Songorica leaf anatomical structure were similar to ellipse,so this dissertation proposed to measure the TKMC and the TVB by the elliptic fitting algorithm.The RANSAC elliptic fitting algorithm and the least square elliptic fitting algorithm were the two commonly used algorithms in elliptic fitting domain at present,the effects of the two algorithms on the edge fitting of kranz mesophyll cell image were compared and analyzed.It showed that R2 were 0.8508 and 0.8979 respectively.Therefore,the least square elliptic fitting algorithm was used to fit the edge of kranz mesophyll cell image to realize the measurement of the TKMC.The measurement was compared with its mean value of interactive measurement,the average relative error of 30 test images was 2.07%and the average time was 1.77s by the measurement proposed in this dissertation.The effects of the above two algorithms on the edge fitting of vascular bundle image were compared and analyzed.It showed that R2 were 0.8753 and 0.9158 respectively.Therefore,the least square elliptic fitting algorithm was used to fit the edge of vascular bundle image to realize the measurement of the TVB.However,the concave points existing between the vascular bundle cells were involved in the least square elliptical fitting caused the poor measurement precision.To avoid this problem,this dissertation proposed a least square elliptical fitting algorithm based on contours convex hull to eliminate the edge concave points of vascular bundles and to realize the accurate measurement for the TVB.The measurement was compared with its mean value of interactive measurement,the average relative error of 30 test images was 1.41%and the average time was 1.92s,while the average relative error was 2.37%by using the least square elliptic fitting algorithm alone.The measurement accuracy of this algorithm was higher than of the least square elliptic fitting algorithm alone.(5)The correlation between leaf anatomical structure indicators and the degradation gradient of grassland was studied.The leaf anatomical structure of Cleistogenes Songorica reacted actively to the gradation of the desert steppe grassland.By single factor analysis of variance,it was found that the thickness of upper epidermis cell(TUE),the thickness of lower epidermis(TLE),the cuticle thickness of upper epidermis(CTUE),the cuticle thickness of lower epidermis(CTLE),the leaf thickness(TL),the thickness of main vein(TMV),the thickness of bulliform cell(TBC),the thickness of mechanical tissue(TMT),and the diameter of vascular bundle(TVB)all increased with the deterioration of the desert steppe degradation,but the leaf width(LW)was decreased.It was also found that the thickness of kranz mesophyll cell(TKMC)did not change significantly different between the control area(CK)and the mild degenerative area(LD),as well as between the moderate degenerated area(MD)and the severe degenerated area(HD);the thickness of vascular bundle sheath cell(TVBSC)changed greatly in the severe degenerated area(HD),which was significantly different from that in the control area(CK)and mild degenerated area(LD).However,there was no significant difference between samples from the control area(LD)and samples from the mild degeneration area(LD).By the correspondence analysis method,it showed that the value of leaf width(LW)was relatively larger in the condition of control(CK)area and slightly degraded(LD)area.The thickness of upper/lower epidermis cells(TUE,TLE),the cuticle thickness of upper/lower epidermis(CTUE,CTLE),the leaf thickness(TL),the thickness of bulliform cell(TBC),the diameter of vascular bundle(TVB)were lager in moderate(MD)area and severe degeneration(HD)area.Different degradation gradients had little effect on the thickness of mechanical tissue(TMT)and the thickness of main vein(TMV).In sum,this dissertation extracted and measured the indicators of the Cleistogenes Songorica leaf anatomical structure by using the image processing technology.It had the advantages of high precision,high efficiency and good reproducibility over other measurements.The dissertation also discussed the correlation between the indicators with high plasticity and the degradation of desert steppe.The stage results obtained in this dissertation can provide a new kind of technical method for the study of plant anatomy,and provide theoretical basis and even technical support for the protection and utilization of Cleistogenes Songorica and the monitoring of desert steppe degradation. |
PDF Full Text Request