In order to satisfy the needs of effective recognition in pest-affected region, a multispectral images acquisition platform was built to monitor the pest-related information efficiently and accurately in Yunnan pine forest region of Yunnan Province. Aneural network of Jeffries-Matusita（J-M）distance optimized back-propagation（BP）neural network was proposed based on unmanned aerial vehicle（UAV）multispectral images. Firstly, the method realized the optimization process of the training samples by introducing the J-M distance concept, which reduced the influence of both “similar spectral from multiple objects” and “multiple spectral from similar objects”. Then, the color and texture features of the images were extracted based on their color and the gray-scale co-occurrence matrix. Three bands of relative spectral reflectance, namely 580Nm, 680Nm and 800Nm were extracted as spectral characteristics. Meantime, five vegetation index models were established to identify pest area. Finally, BP neural network algorithm was applied for training and identifying four feature vector quantities, including color, texture, spectral and vegetation index, which greatly achieved the identification and classification goal of pest region. The proposed algorithm was compared with the traditional BP neural network and support vector machine（SVM）algorithm from both general classification precision and the Kappa index. The experimental results showed that the overall accuracy index of classification and the Kappa index of the algorithm reached 94.01% and 0.92, respectively, which was superior to traditional BP neural network and SVM algorithm. Besides the modeling time was shortened by 38% when compared with the traditional BP neural network method. The improved efficiency satisfied the high efficiency identification needs of Yunnan pine pest area in Xiangyun County.