The machine vision technique is introduced to recognize the Empoasca flavescens automatically on the yellow sticky traps in natural scenes in order to realize the accurate and timely forecast of Empoasca flavescens in tea garden. The superpixel segmentation algorithm and DBSCAN (density-based spatial clustering of applications with noise) cluster algorithm are employed to separate the interesting target regions from background, which ensures the accuracy and completeness of the target area. Then, six classification features, including mean value of L, a, and b and their standard deviation, are extracted from the marked area in target image. Last, LSSVM (least squares support vector machine) is developed to identify Empoasca flavescens from other insects that are captured by sticky traps. As the imbalanced sample number between Empoasca flavescens and other insects results in the low classification accuracy, the improved SMOTE (synthetic minority over-sampling technique) algorithm and KS (Kennard-Stone) algorithm are used to improve recognition accuracy of Emposace flavescens. The proposed algorithm achieves 99.03% of the overall recognition accuracy, and the identification accuracy of Empoasca flavescens reaches 91.76%. The proposed algorithm can provide an effective way for real-time detection of Empoasca flavescens.