IMNM： Integrated Multi-Network Model for Identifying Pepper Leaf Diseases

Zhaopeng Cai^1,2*Nadia Farhana^3*Asif Mahbub Karim^4*Fengyan Zhai⁵Wenwen Huang⁶Meng Guo^1,2

• ¹Research Center of Smart City and Big Data Engineering of Henan Province, Research Center of Intelligent Campus Application Engineering of Henan Province, Henan University of Urban Construction, Pingdingshan, China
• ²School of Computer and Data Science, Henan University of Urban Construction, Pingdingshan, Henan Province, China
• ³Department of Business Administration, Stamford University Bangladesh, Dhaka, Bangladesh
• ⁴Binary Graduate School, Binary University of Management & Entrepreneurship, Puchong Selangor, Malaysia
• ⁵Department of Plant Pathology, Henan Province Engineering Research Center of Biological Pesticide & Fertilizer Development and Synergistic Application, Henan Institute of Science and Technology, Xinxiang, China
• ⁶School of Resources and Environment Science, Henan Institute of Science and Technology, Xinxiang, Henan Province, China

Pepper is one of the most valuable vegetable crops. However, pepper leaf diseases have a significant impact on the growth of pepper yield. Since the spots of these diseases were complex in color and texture. Yet, their manual prevention and identification require much time and energy. It is critical to design a more efficient pepper leaf disease identification framework for improving the yield and quality of pepper. This study combines an improved ResNet, a dynamic convolution network (DCN), and a progressive prototype network (PPN) to create an integrated multi-network model (IMNM). This study concentrated on five different types of pepper leaf samples: healthy, virus, leaf blight, brown spot, and phyllosticta.Experimental results demonstrated that the IMNM yielded an identification accuracy score of 98.55%, superior to the outstanding Inception-V4, ShuffleNet-V3, and Efficient-B7. In addition, more experiments were carried out to confirm its generalization capabilities. The IMNM was used to identify the apple, wheat, and rice leaf diseases and achieved an average identification accuracy of 99.81%, and specificity, precision, sensitivity, and accuracy, all four indicators are all much greater than 98%. It has been confirmed that the IMNM can identify many agricultural leaf diseases, demonstrating its strong generalization capabilities. The proposed technique is a suitable starting point for promoting the intelligent identification of crop diseases and pests, and it includes suggestions for transferring deep learning models to field disease mobile identification equipment.