Pest infestation reduces fruit quality and causes economic losses. Accurate identification of citrus pests is conducive to pest control. However， as the features to distinguish these pests are not obvious， manual classification is time-consuming and labor-intense， while advanced algorithms have high computational costs. Therefore， it is necessary to develop lightweight and accurate identification tools. In this article， a data set of insect pest images containing 10 types of pest images that are most harmful to citrus was constructed. A network featuring lightweight and high precision was developed based on MobileNet-V2 and the attention mechanism ECA. Moreover， an edge computing APP was also developed that can be run on Android phones. The ECA attention mechanism was embedded in the tail of the anti-residual structure of the improved MobileNetV2 network to enhance the cross-channel information interaction ability and improve the feature extraction ability. The results of testing showed that the ECA_MobileNetV2 model had a classification accuracy of 93.63% for citrus pests， 1.68， 1.44 and 2.40 percentages higher than that of the MobileNetV2， GoogLeNet and ResNet18 models， respectively. The parameter， FLOPS and size of model was 3.50×10⁶， 328.06×10⁶ and 8.72 MB， respectively. Its complexity is only slightly higher than that of MobileNetV2， and it can run in the form of edge computing on mobile phones. It is indicated that the developed intelligent recognition tool can quickly and effectively classify and identify different types of citrus pests.