Ensemble Deep Learning Applied in Precision Farming Through Controlled Fertilisation, Disease Mitigation, Optimised Harvesting and Marketing

Abstract

Agriculture is considered as largest economy in 
the world especially in India as it provides huge no of 
employment and nutrient benefits of various crop and plant for 
human survival. However, due to reduced no of research in 
agriculture field has impacted the farmer on basis of crop yield. 
In order to enhance the yield of crop, various researches has 
been conducted in past decade especially in terms of detecting 
the crop diseases. In specific, it becomes mandatory to analyze 
the suitable crop to specific soil condition and environmental 
condition. Meanwhile application of fertilizer is significant to 
increase the soil condition for crop cultivation and increase crop 
yield, Thus, a ensemble deep learning architecture is designed 
and implemented with functionalities of crop recommendation. 
Plant disease identification is becoming significant challenge all 
over the world, therefore automatic detection and arrangement 
of the disease of plant leaf is very important in monitoring the 
plant. In this paper, ecommerce application is developed using 
ResNet architecture towards plant disease prediction. 
Additionally, aggregation of Random Forest algorithm and 
Support vector Machine is used for fertilizer recommendation 
on basis of soil and environmental conditions. Convolution 
neural network is modeled to classify the leaf image into 
multiple disease classes. However convolution layer and max 
pooling layer of the architecture produces the features map in 
order to increases the classification efficiency. Further ReLu 
activation function is employed in fully connected layer to avoid 
the over fitting issues to enhance the scalability and accuracy of 
the model on classify feature map into various diseases classes 
of the plant leafs as leaf blight, gray mold, powdery mildew. 
Experimental analysis of the model is carried out using plantvillage dataset. Finally efficiency of the model is evaluated on 
basis of accuracy and loss to the training and validation data.