Identification of Pests and Diseases in Greenhouse Rice Based on ConvNeXt-T Neural Network

Rice cultivation in greenhouses is a key agricultural form in our agricultural development. Timely detection and prevention of pests and diseases in rice in greenhouses has a significant impact on improving rice yield. Deep learning models excel in image recognition and can be used to monitor rice-induced growth conditions in rice in greenhouses and quickly identify diseases and pests. Image data analysis enables farmers to take timely measures to prevent the spread of pests and diseases. Various environmental factors affect the collected dataset, resulting in an insufficient number of available images, and the training process can easily fail to extract effective features. Addressing these problems, this paper proposes a rice pest and disease recognition model based on the ConvNeXt-T neural network. Data enhancement techniques, such as mirroring and cropping, and data preprocessing steps, including the addition of Gaussian noise, random brightness, and random masking, were applied to the dataset. The initially acquired 5,932 rice pest images were expanded to 21,340 images. These augmented images were then used to train a ConvNeXt-T neural network model for recognizing four of the most common diseases of rice: leaf blight, rice bacterial streak, brown mottle, and rice dong quai virus disease. The experimental results demonstrate that the ConvNeXt-T neural network performs optimally, achieving the highest level of disease recognition accuracy (99.86%) compared to the classical AlexNet, GoogLeNet, ResNet34, and VGG16 networks in the same experimental environment. Its excellent recognition accuracy provides strong support for the prevention of pests and diseases in greenhouse riceacterial streak, brown mottle, and rice dong quai virus disease). The experimental results show that the ConvNeXt-T neural network performs optimally and achieves the highest level of disease recognition accuracy (99.86%) compared with the classical AlexNet, GoogLeNet, ResNet34 and VGG16 networks in the same experimental environment. Its excellent recognition accuracy provides strong support for the prevention of pests and diseases in greenhouse rice.