An Enhanced Deep Learning Methodology for Identifying Diseases Pertaining to Maize Crop using Improved Gaussian Particle Swarm Optimization

Jayapriya Palanisamy*

Department of Computer Science, Nallamuthu Gounder Mahalingam College, Pollachi, Tamil Nadu, India.

Corresponding Author E-mail: topriyadamu@gmail.com

DOI : http://dx.doi.org/10.12944/CARJ.12.2.18

Abstract:

Image processing is a method of transforming images into digital form to carry out contributory techniques for the segmentation and classification of images. It is a type of processing that relies on automatic and accurate identification for quantifying disease. The significant growth of India’s economy has been partially based on maize-related industries. To maintain the prosperity of the maize industry, it is important to address disease control, labour cost, and global market concerns. In recent years, Leaf Blight, Leaf Spot, and Common Rust have become severe threats to maize plants in India. These diseases can result in plant death, loss of yield, and marketability loss This research proposes three new compute-intensive approaches, namely Improved Deep Multiscale Convolution Neural Network (IDMCNN), Enhanced Bacterial Foraging Optimized Recurrent Neural Network (EBFORNN), and Improved Gaussian Particle Swarm Optimized Convolution Neural Network (IGPSOCNN).The above approaches, IDMCNN, EBFORNN, and IGPSOCNN, consist of three major phases: Image processing, Segmentation, and Classification. In the first part of the proposed work, the Improved Deep Multiscale Convolutional Neural Network (IDMSCNN) uses Sparse Principal Component Analysis (SPCA) and Affinity Propagation (AP) for image preprocessing. Deep Multiscaling is employed for maize disease segmentation, and classifications are performed using pooling, activation functions, and a saliency map. In the second part of the work, the Enhanced Bacterial Foraging Optimized Recurrent Neural Network (EBFORNN) segments maize plant diseases using Bacterial Foraging (BF) and classifies them using an optimized Recurrent Neural Network (RNN). The Recurrent Neural Networks of EBFORNN have higher recognition accuracy than IDMSCNN. The third work, the IGPSOCNN, handles segmentation and classification using Gaussian particles with a Convolutional Neural Network (CNN). These concepts are engaged to improve segmentation and offer better classification in the network. Initially, Contrast Limited Adaptive Histogram Equalization (CLAHE) is used for preprocessing. The Color Concurrence Matrix (CCM) is emphasised to generate optimal features through feature extraction.

Keywords:

Bacterial Foraging; Convolution Neural Network; Deep multiscale; Particle Swarm Optimization; Recurrent Neural Network

Download this article as: 

[ HTML Full Text]