Apple Leaf Disease Identification Based on Optimized Deep Neural Network

Introduction

Apple trees are mostly affected by pests and diseases that are caused by bacterial and viral infections which lead to the reduction in apple cultivation. Diseases affecting apple trees are of great concern, since they affect the productivity and quality to a large extent. Apple diseases detected so far exceeds more than hundred in number, increase in apple diseases affects the apple production to a great extent. The three common type of disease which affects the apple leaves which are considered in this work are Black Rot, Rust and Apple Scap. The apple diseases can be controlled, only if it is properly identified (Arora & Singh, 2019). This work concentrates on identifying the diseased leaves rather than the tree. Manual identification of the diseased apple leaves is incompetent and costly .The pattern change in apple leaves due to the disease is very minor, this minor variation cannot be effectively monitored by human beings. Computer Vision techniques are efficient in detecting the apple leaf diseases effectively. There is a color and texture change in the affected leaves which are different from normal leaves that can be detected to identify the disease. Many technologies have emerged by considering the colour, texture and shape as differentiating factors to diagnose the plant disease with a threshold between normal and pathological sections using a single feature. The existing technologies use single feature which is not effective. Conventionally the disease diagnosis in the apple leaves are done using machine learning techniques such as k-Nearest Neighbour, Random Forest, and Support Vector Machine, with enhanced recognition rate. The recognition rate is less and is still susceptible. Deep Convolution Neural Network approach which has developed in the recent years is an end-to-end pipeline which automatically determines the discriminative features for image classification. Convolution Neural Network is considered as one of the most excellent classification technique for pattern recognition task. Stimulated with the development of the convolution neural networks in image based recognition, the need of CNN in identifying early diseases has been a new research in recognition of diseases in apple trees (Prasad et al., 2019). This work uses CNN for feature extraction. Developed in the recent years CNNs are highly useful in the field of disease recognition for crops. This work proposes a CNN based approach for identifying the disease features from the affected apple leaves. The investigation of Convolution Neural Networks has not only decreased the need for image pre-processing, but also improves the detection accuracy. The proposed work uses more than thirty features of colour, shape and texture. At first, the input leaf disease image is pre-processed. In pre-processing stage, the quality of the input image is enhanced. Then the resultant pre-processing output is fed to the feature extraction process. For that, the suggested method utilizes the Convolutional Neural Network (CNN). A Convolutional Neural Network is a type of Deep Neural Network, normally applied for analyzing visual images. An Artificial Neural Network (ANN) with several layers between the output and input layers is termed as the Deep Neural Network (DNN). The DNN turns the input into output by finding the correct mathematical manipulation, even if the relationship is a linear one or a non-linear one. The effective leaf infection detection is done with the support of ODNN. As of late, meta-heuristic algorithms have proven to be efficient when compared to the other optimization algorithms. A meta-heuristic algorithm solves the conventional optimization problems such as Particle Swarm Optimization, Genetic Algorithms etc. Generally, meta-heuristic optimization methods aids in providing a suitable solution by trial and error method in a given exact designated time. Here the traditional DNN algorithm is modified by means of weight optimization using Adaptive Monarch Butterfly Optimization (AMBO) algorithm. The Monarch Butterfly Optimization (MBO) mimics the immigration behaviour of the monarch butterflies, to solve global optimization problems. The majority of metaheuristic algorithms, neglects the process of updating the information available from the previous iterations .The proposed Adaptive Monarch Butterfly Optimization algorithm initialize an initial solution and an opposite solution is obtained for both lower and upper values .This improves the optimization process significantly .To improve the accuracy most important features are selected using the Adaptive Monarch Butterfly Optimization (AMBO) algorithm (Arora & Singh, 2019), and disease detection is done with Optimal Deep Neural Networks(ODNN). A Deep Neural Network (DNN) is a type of Artificial Neural Networks with several connections linking the input and output nodes form a directed graph next to a temporal sequence. Bio-inspired meta-heuristics perform better in selecting better features than the traditional approaches. This chapter uses the Adaptive Monarch Butterfly Optimization (AMBO) algorithm which is a bio-inspired algorithm that mimics the migration behaviour of the monarch butterflies, to solve the feature optimization problem (Ibrahim & Tawhid, 2019). The weight optimization in the traditional DNN algorithm is done with the help of the Adaptive Monarch Butterfly Optimization (AMBO) algorithm. Optimal features are selected by employing AMBO and disease identification is done with the help of ODNN. In this research, efficient apple leaf disease identification is done by Optimal Deep Learning Approach. Performance of the proposed technique is estimated by accuracy, sensitivity and specificity. The implementation work of the proposed method has been done in MATLAB. From the results it is evident that the proposed method proves to be a better technique than the existing methodologies for disease identification