An LSTM Based CNN Model for Pomegranate Fruit Classification With Weight Optimization Using Dragonfly Technique

Abstract

Pomegranate is a widely grown plant in India. This highly beneficial fruit is infected by multiple pests and diseases which cause great economical losses. Different forms of pathogen diseases on leaf, stem and the fruits are present. Some of the diseases that affect pomegranate fruits are anthracnose, cercospora, heart rot and bacterial blight. There is a need for disease control strategies to incorporate timely action on the developed diseases. Thus, there is a need for intelligent and self-learning recognition systems to detect these diseases on time. This study is aiming to classify pomegranate fruits into two classes normal and abnormal using CNN LSTM technique. This research work uses a hybrid CNN-LSTM technique to detect four types of diseases present in the pomegranate fruits and classify them into four classes. The results obtained using CNN LSTM are then optimized using dragonfly algorithm. The features like colour, texture and shape of the fruits are collected and fed into the hybrid CNN-LSTM. The dataset for the classifier is given as an excel file which is initially pre-processed using map reduce technique and dimensionality reduction carried using Principal Component Analysis and Discriminant analysis. The CNN LSTM classifier identifies the 4 types of diseases and normal fruit. The classification is further optimized using dragonfly algorithm. The optimized weight and cost function has further explored to support the multi-class disease detection process. Experimental results have shown an accuracy of 92% in classification using CNN-LSTM technique and optimization using dragonfly techniques shows an improved classification accuracy of 97.1%.

Bayesian Network Structure Learning Algorithm Combining Improved Dragonfly Optimization

“…The experiment showed that this method had a maximum signalto-noise ratio of 52.8dB and a minimum error of only 0.19. Vasumathi M T and his team utilized the dragonfly algorithm to optimize the detection performance of long and short term convolutional networks, and applied the resulting hybrid algorithm to the identification of pests and diseases in pomegranate plants [11]. According to the test results, the classification accuracy of using long and short term convolutional networks alone was lower than that of networks optimized by the dragonfly algorithm, with the latter achieving a maximum accuracy of 99.1%.…”

“…In addition to illegal structures, the attenuation of algorithms is also a problem that needs to be considered in BNSL.In the DO algorithm, the variation of its inertia weight w follows a linear function. Assuming max t as the maximum number of iterations, the mathematical expression of w is shown in formula (11). max (0.90.4) 0.9 t w t…”

Fruit Image Classification Using Deep Learning Algorithm: Systematic Literature Review (Slr)

“…Vasumathi dan Kamarasan,2021, melakukan penelitian dengan model CNN berbasis LSTM pada mengklasifikasikan buah delima menjadi dua kelas normal dan abnormal dengan optimasisasi Teknik capung, hasil eksperimen menunjukkan akurasi 92% dalam klasifikasi menggunakan teknik CNN-LSTM dan optimasi menggunakan teknik capung menunjukkan peningkatan akurasi klasifikasi sebesar 97,1% [15].…”