基于颜色和纹理特征的玉米旱情统计判别模型

利用颜色和纹理特征判定玉米植株干旱程度是玉米旱情监测识别的新途径,良好的统计判别模型构建方法对实现这一新途径意义重大。采集了玉米出苗-拔节、拔节-抽雄、抽雄-成熟3个生长发育阶段适宜、轻旱、中旱、重旱、特旱5个干旱程度的玉米植株图像,用MATLAB软件从图像中提取玉米植株的颜色和纹理特征数据,以SPSS软件对特征数据的训练集进行Fisher逐步判别分析,获得Fisher判别函数组,结合曼哈顿距离判别规则,建立了玉米出苗-拔节、拔节-抽雄、抽雄-成熟3个生长发育阶段的正视面、俯视面和侧视面共9个单视角统计判别模型,并将每个生长发育阶段的单视角统计判别模型联合构建为三维统计判别模型。玉米3个生长发育阶段的所有单视角判别模型在训练和测试时的平均判别准确率无显著差异,所有正视面和侧视面单视角统计判别模型训练和测试时的平均判别准确率均在90%以上,判别玉米不同干旱程度的准确率差异性小,所有俯视面单视角统计判别模型训练、测试时的平均判别准确率均低于85%,且判别玉米不同干旱程度的准确率差异性较大。玉米3个生长发育阶段的三维统计判别模型的训练和测试平均判别准确率在97%以上,且判别玉米不同干旱程度的差异性较小。玉米3个生长发育阶段的单视角统计判模型中,侧视面统计判别模型的判别效果最好,正视面统计判别模型次之,俯视面统计判别模型的判别效果相对较差,每个生长发育阶段的三维统计判别模型的判别效果均比单视角统计判模型好。

Maize Drought Statistic Discriminant Model Based on Color and Texture Features

Using color and texture features to discriminate drought degree of maize plants is a new way for monitoring and identifying maize drought damage. It is of great significance to establish a good statistical discriminant model to realize that new way. This study collected images of maize plants under 5 drought degrees of free, light, medium, severe and extreme drought stress during 3 growth and development periods: emergence-jointing, jointing-tasselling and tasselling-maturing; and extracted the color and texture feature data of maize plants from the images by MATLAB software. Then, Fisher discriminant function groups were acquired through Fisher step by step discriminant analysis of the feature data training set by SPSS software. The 9 single-view(frontal, top or lateral view) statistical discriminant models for drought maize plants in emergence-jointing, jointing-tasselling and tasselling-maturing 3 growth and development periods were established by combining Fisher discriminant function groups and Manhattan distance discriminant rule. The single-view statistical discriminant models for drought maize plants in each growth stage were jointly constructed as a 3D statistical discriminant model. There was no significant difference in average discrimination accuracy between training and testing of all the single-view discriminant models of 3 maize growth stages. The average discrimination accuracies at training and testing of frontal and lateral view statistic discriminant models were over 90%. The difference of discriminating the different drought degree of frontal and lateral view statistic discriminant models maize was small. The average discrimination accuracies of all top view statistical models at training and testing were less than 85%, and there were significant differences in discriminating different drought degrees of maize. The accuracies of 3-dimensional statistical discriminant models for drought maize plants of 3 growth stages at training and testing were all over 97%, and had little difference in discriminating different maize drought degree. Among the single-angle statistical discriminant models, the discriminant effects of lateral statistical discriminant models were proved to be the best, followed by frontal statistics discriminant models, and the discriminant effect of the top view discriminant model was relatively poorer. The discriminant effect of 3D statistical discriminant model for maize plants in every growth stage was better than that of single-angle statistical discriminant models.