The Study On The Detection Of Grape Leaves And The Measurementmethod Of Leaf Area

Recent years,with the mode of agriculture production become more and more intelligent,meticulous and mechanical,application of video monitoring in agriculture will be more and more extensive and widespread. As the primary monitoring object, the plant leaves directly reflect plant diseases and growth state.The paper researches the method of automatic detection of grape leaves under natural enviroment with HOG(Gradient direction histogram) descriptor and SVM(Support vector machine).The method includes two steps,the recognition and the localization, through the identification phase to determine the existence of the target leaves, through the localization phase to finally determine the position of the target leaf.First of all, histogram of oriented gradient(HOG) features of a leaf are extracted,and a classifier is trained by support vector machine to identify the grape leaves in the images. The experiment results show that the method combing HOG features with Support vector machine can effectively reduce the interference of uneven illumination and background changes,also it has a high recognition rate under natural conditions.Because the position of leaf in the image is arbitrary,the size and number of leaves is unfixed,hence,during the localization phase,we scale the images in different scales to get a series of size reduced image sets,then extract each image’s HOG features using a sliding,size fixed window to judge,eventually realize the localization of grape leaves with non maxima suppression algorithm.The experiment results show the method can realize on-line detection of grapeleaves under natural conditions, the detection rate of grapeleaves of the images under natural conditions is over 80% in experiment,it has strong robustness to illumination and environment change.In addition,an improved corner detection algorithm combining SUSAN and Shi-Tomasi algorithm is proposed in this paper,the area of grape leaf is measured accurately according to the improved corner detection algorithm and perspective transformation principle. Experimental results show that the method is effective for reducing the measurement error due to axis of imaging equipment is not vertical to leaf。...