Abstract: Maize, as one of the world's major food crops, faces a number of serious threats from various and awful diseases. These crop diseases, such as Leaf spot disease, Gray spot disease, Mosaic disease, and Rust disease, can lead to significant yield losses and quality degradation of maize, thus becoming a crucial constraint to maize production development. In the field of crop disease identification, YOLOv8 has gained widespread recognition as a representative lightweight model due to its excellent balance between accuracy and computational efficiency. However, despite its advantages, YOLOv8 still has room for improvement when it comes to practical detection deployment, especially in resource-constrained environments. To address issues such as the severe impact of corn diseases represented by leaf spot disease, gray spot disease, mosaic disease, and rust disease on corn yield and quality, as well as to enable real-time detection of corn diseases on edge computing devices, this paper proposes a lightweight, high-precision model called YOLOv8-Corn.First of all, This study uses YOLOv8n as the baseline model, first adding deformable convolutional networks (DCNv2) and the DySample (dynamic sample) upsampling operator module to improve accuracy. Then, we proposed the use of Network thinning and pruning techniques, which allowed us to remove approximately 25% of the redundant parameters from the model. Not only did this process reduce the model size but it also maintained its essential detection capabilities. After obtaining a relatively compact pruning model, we further employed callback training. This training strategy aims to supplement the weight removed by pruning, thereby restoring and even improving the accuracy of the model to a certain extent.Subsequently, we combined the model with the CWD(Channel-Wise Knowledge Distillation) feature distillation method. This step is of great significance as it enables further improvement of the model accuracy while ensuring lossless reduction of model parameters and calculation amount. Last but not the least, with the intention of achieving practical deployment, we integrated the lightweight model with TensorRT and completed the actual deployment on the Jetson Nano platform.The results of this study, in addition to their promising characteristics, demonstrate the effectiveness of our approach. Compared with other networks, the improved model of YOLOv8n shows superior performance in terms of detection accuracy and lightweight level. Specifically, the average accuracy rate, recall rate as well as average accuracy of the improved model reach 99.78%, 99.30%, and 99.20%, respectively. These high indicators highly indicate that the model can accurately identify maize diseases in practical applications. Moreover, the model size is reduced from 6.35 MB to 4.40 MB, making it more suitable for deployment on devices with limited resources. The number of frames per second reaches 121.3, ensuring real-time performance during detection. By realizing the joint with Jetson Nano, we have successfully completed the actual deployment and application of the detection model.In conclusion, the research presents a highly efficient and lightweight detection model for maize diseases. The designed model and the actual deployment of the network provide strong technical support for the realization of lightweight and high-precision detection of maize underleaf diseases. Apart from its conspicuously contribution to the advancement of crop disease identification technology, this work boasts practical significance for improving maize production and quality. The model's high accuracy and lightweight characteristics make it a valuable tool for real-time monitoring and management of maize diseases, effectively facilitating in a further step offering a promising solution for precision agriculture and sustainable crop production. Moreover, this research addresses a critical need in agricultural technology by providing a robust and efficient solution for maize pest and disease detection. Through the seamless integration of the combination of advanced model optimization techniques and practical deployment strategies makes this study a truly valuable contribution to the field of agricultural intelligent equipment engineering.