Feature Extraction Research Of Maize Leaf Drought Resistance Based On Image Processing

In the current situation of water resource shortage, the studies about maize drought resistance have great significance for the development of water saving agriculture and increasing maize yield. The traditional maize drought detection mainly rely on breeders manual observation and direct field identification, this process could be low efficiency and error prone. In this study, the digital image processing applied to maize drought resistance detection, at first, acquiring visible images and infrared images of maize leaves in the crop filed, studying on image processing and feature extraction methods, extracting the morphological characteristics, color characteristics and temperature characteristics. Next, analyzing the relationship between these characteristics and conventional maize drought resistance indexes, such as chlorophyll value and stomatal conductance. Finally, extracting the leaf characteristics which can reflect the maize drought resistance. The main research contents and conclusions of this study are as follows:(1) The morphological characteristics extraction of maize leaf drought resistances. Through carrying image processing operation on visible image of leaves, such as gray-scale, segmentation, morphological, extracting the morphological characteristics of Leaf Rolling Index(LRI) and per plant Coverage. Through comparing the effects of different threshold segmentation methods on per plant Coverage and found that after Super-green’s Method segmentation, the per plant Coverage is the most close to the real value. Through analyzing the differences of Leaf Rolling Index(LRI) and per plant Coverage among maize varieties with different drought resistance, the results showed that the two maize leaf morphological characteristics of LRI and per plant Coverage could reflect the maize drought resistances.(2) The color characteristics extraction of maize leaf drought resistances. Through acquiring the flag leaf visible images of 9 different drought resistant maize varieties in seedling, extracting the 21 color features in RGB and HSI color space and implanting the correlation analyze between 21 color features and the chlorophyll values, the results show that there is a significant correlation between G, g, H, R/G, R/B, R/G+B, G/B, B/R, 2G-R 9 color features and chlorophyll values. Through using principal component analysis(PCA) for 9 multiple color features, the results show that the cumulative contribution of the former 4 principal components after dimensionality reduction is 99.6567%.(3) The temperature characteristics extraction of maize leaf drought resistances. Through comparative analysis of different enhancement algorithms and segmentation algorithms of the thermal infrared image, the results show that using Homomorphic Filter Algorithm to per plant maize canopy thermal infrared images is more conducive for image segmentation, using the segmentation algorithm based on the combination of K-means and Otsu could achieve effective segmentation. Through using thermal infrared image segmentation method extract the temperature characteristics of the per plant canopy leaf region, and analyzing the relationship between per plant canopy leaf region temperature characteristics and conventional maize drought resistance indicators, the results showed that the temperature difference of per plant canopy leaf region between morning and afternoon could reflect differences of maize varieties among different drought resistance.