Abstract: Aiming at the problems of poor real-time performance, low detection accuracy, and over-reliance on high-computing and high-power hardware for identifying Clivia miniata pests and diseases in greenhouse and garden environments, Proposing a YOLO V4-TLite Clivia miniata pest detection method for mobile execution. Firstly, the dataset of Clivia diseases and insect pests comes from the real greenhouse planting environment. The picture collection time is winter and spring. The degree of disease includes the early and middle stages of Clivia diseases and insect pests. Secondly, based on the YOLO V4-Tiny model, a low-cost improved partial convolution is used to replace the traditional convolution in the backbone network, so that the improved model can have faster operation speed and memory consumption. Thirdly, aiming at the problem of hardware compatibility caused by the large random storage consumption in the backbone network of the YOLO V4-Tiny model, an improved partial convolution and inverse residual network structure is used to form a lightweight backbone network, so that the improved model does not have large random storage consumption in the depth of the backbone network, which improves the operation speed of the model and the compatibility of the mobile terminal with limited resources. fourthly, given the problem that the traditional convolution layer of the YOLO V4-Tiny model has more redundant feature maps and feature map attention distraction, the weight sharing convolution and conventional convolution are used for channel fusion operation, so as to improve the robustness and accuracy of the network. Finally, the improved model migration was deployed on the ROCK 5B mobile and tested against three typical Clivia miniata pests: leaf blight, maculopathy, and coccid. The experimental results showed that the mean average precision (mAP) of 78.5% at an intersection over union (IoU) ratio of 0.5 for this improved model, with a memory usage of only 4.8MB, and the floating point operations (FLOPs) is 1.3 G. The desktop single detection speed is 0.005 s with 70 W power consumption. On the mobile side, the CPU single detection speed is 0.239 s with 10 W power consumption, and the NPU single detection speed is 0.018 s with 7 W power consumption. Compared with the original YOLO V4-Tiny model, the mAP₅₀ of the YOLO V4-TLite model increased by 12.6 %, and the model size decreased by 78.6 %. Compared with the YOLO V4-Tiny model, the computational efficiency of the YOLO V4-TLite model is improved by 35.7 % and the power consumption demand is reduced by 26 W on the desktop side. The computational efficiency on the NPU side of the mobile side is also improved by 85.7 % and the power consumption demand is reduced by 2 W. Compared with other backbone networks MobileNetV2 and GhostNetV1, its mAP₅₀ is 3.7 % and 5.5 % higher, respectively. Compared with the target detection models YOLOV11-N, YOLO V10-N, YOLO V7-Tiny and YOLO V5-S, the mAP₅₀ is 3.9 %, 2.3 %, 1.6 % and 1.3 % higher, respectively. The YOLO V4-TLite model performs Clivia Miniata pest and disease detection in a low-resource and low-power mobile, and it is competitive with the existing mainstream YOLO series models.