Internet of things based smart application for rice leaf disease classification using optimization integrated deep maxout network

被引:2
|
作者
Shanmugam, Vimala [1 ]
Rao, Telu Venkata Madhusudhana [2 ]
Rao, Hanumantu Joga [3 ]
Maram, Balajee [4 ]
机构
[1] Prathyusha Engn Coll, Dept Elect & Commun Engn, Chennai, Tamilnadu, India
[2] Vignans Inst Informat Technol, Visakhapatnam, India
[3] GMR Inst Technol, Dept Chem Engn, Rajam, Andhra Pradesh, India
[4] Chitkara Univ, Sch Engn & Technol, Baddi, Himachal Prades, India
来源
CONCURRENCY AND COMPUTATION-PRACTICE & EXPERIENCE | 2023年 / 35卷 / 06期
关键词
data augmentation; deep maxout network; Gaussian filtering; internet of things; plant disease detection; routing;
D O I
10.1002/cpe.7545
中图分类号
TP31 [计算机软件];
学科分类号
081202 ; 0835 ;
摘要
Rice is the major crop in India. Early prevention and timely identification of plant leaf diseases are important for increasing production. Hence, an effective sunflower earthworm algorithm and student psychology based optimization (SEWA-SPBO) based deep maxout network is developed to classify different types of diseases in rice plant leaf. The SEWA is the combination of sunflower optimization (SFO) and earthworm algorithm (EWA). Initially, the network nodes simulated in the environment capture the plant leaf images and are routed to the sink node for disease classification. After receiving the plant images at the sink node, the image is preprocessed using a Gaussian filter. Next to preprocessing, segmentation using the black hole entropic fuzzy clustering (BHEFC) mechanism is performed. Then, data augmentation is applied to segmented image results and disease classification is done by a deep maxout network. The training of the deep maxout network is done using the proposed SEWA-SPBO algorithm. The proposed method detects the leaf disease more accurately with limited time and shows higher accuracy. Moreover, the proposed method attains higher performance with metrics, like accuracy, sensitivity, and specificity as 93.626%, 94.626%, and 90.431%, respectively.
引用
收藏
页数:21
相关论文
共 50 条