ass日本风韵熟妇pics男人扒开女人屁屁桶到爽|扒开胸露出奶头亲吻视频|邻居少妇的诱惑|人人妻在线播放|日日摸夜夜摸狠狠摸婷婷|制服 丝袜 人妻|激情熟妇中文字幕|看黄色欧美特一级|日本av人妻系列|高潮对白av,丰满岳妇乱熟妇之荡,日本丰满熟妇乱又伦,日韩欧美一区二区三区在线

Early blight disease is a common disease of greenhouse tomato, which seriously damages the yield and economic benefits. As affected by complex background such as soil, ground, plastic film and lots of overlapping green leaves in greenhouse, it is difficult to recognize disease from image of tomato leaf. In order to provide a solution for such problem, an innovate tomato early blight disease spot detection method of sliding window SVM (SW-SVM) was proposed. To enhance recognition accuracy and stability, color and texture features included color moment (CM), color coherence vector (CCV) and rotation invariant co-occurrence among adjacent LBPs (RIC-LBP) features were introduced, and CCR-SVM (CM+CCV+RIC-LBP+SVM) classification model were trained by using RBF-SVM with the extracted color texture feature (CCR) from the training samples. Meanwhile, for supporting small region data set and to fulfill recognize performance under complex environment, original images were divided to small region images by applying sliding window. And small region images belonged to early blight disease spot, healthy leaves and ground background were selected and divided into three catalogs as training samples. To verify feasibility of the proposed method, offline and online experiments were conducted. For offline classification performance, cross validation average recognition rate was 99.55% and recognition rate for testing data set was 96.97%, and average testing time for a single sliding window image was 0.004s. For online detection performance, the results showed that the proposed method can realize average accuracy rate for the original images with 86.39%, average detection time of single sliding windows image with 0.073s. For rotated images and pixel value adjusted image data, average accuracy rate was 88.98% and 92.59%, respectively; average error recognition rate was 12.71% and 16.44%, respectively; average missing recognition rate was 10.93% and 7.41%, respectively; and average disease detection time of single sliding window image was 0.075s and 0.074s, respectively. As a conclusion, the offline and online experiments results showed that the proposed method of CCR-SVM realized high accuracy and low memory requirement, which could provide real-time solution for tomato early blight detection in greenhouse.