Deep Learning and IoT: The Enabling Technologies Towards Smart Farming

Introduction

The population of the world is growing rapidly, and the demand of the food is increasing. The world is using modern techniques of smart farming to increase the crop productivity with the help of IOT, cloud computing, machine learning and deep learning. Deep learning is the sub field of AI and machine learning. In order to meet large increasing demand of food the food security and crop yield prediction is important. This can be achieved by using the technologies such as deep learning. Crop diseases cause huge damage to the crop. The disease affects different areas such as leaf, flower, stem etc. The disease identification at an early stage is difficult, time consuming and costly. Due to multiple diseases in the crop, the crop does not give enough production and does not meet the over all demand of the food. The crop disease identification with manual methods such hiring experts to visit the land physically is a very difficult job to visit whole land and identify the infected areas. As human can overlook and leave the infected areas unattended. This does not help to asses the crop correctly. Therefore, the technology can help in this area to assess the disease early when it first appears. The correct and timely disease identification helps farmers and agronomists to protect the crop from the loss, to increase the crop production and to meet the food demand. The deep learning is a new field of AI which helps a lot to identify the crop diseases timely and effectively. The correct disease identification (Kamilaris and Prenafeta-Boldú, 2018) is a difficult job. First step in disease identification is to capture image through mobile camera, drone camera or through different sensing devices such employing different camera sensors, then analyzing and classifying the images according to diseases they contain. The sensing methods are used to acquire images remotely includes satellite, multispectral, hyperspectral, near infrared and many more. There are many popular techniques available to process digital image and classify image according to the infected parts of the leaf, fruit or stem. The commonly used techniques for analyzing and classifying images includes machine learning, K-Means algorithm and SVM. DL is like ANN. These all mentioned techniques are good at some point but when we it is required to know deep inside then the Deep Learning is more powerful technique to solve this type of problem.

Deep Learning

Artificial Intelligence is a broad field and has been for long time. The deep learning is the sub field of Artificial Intelligence. The human brain contains almost 86 billion neurons. The activity of human spinal card and other things are carried out by neurons in the brain. The Natural Neuron contains a very complex structure. The Artificial Neural Networks mimics the Natural Neuron. The artificial neurons work same as natural neurons where one neuron carries information and passes to another neuron. The deep learning deals with neural networks with more than two layers. Deep learning methods (LeCun, Bengio, and Hinton, 2015) are also called representation learning methods. They give multiple level of representation. We take an example of image as we know image comprised of pixel values and represented in the form of array. The first layer typically represents the presence or absence of edges at some location in the image. The second layer detects motifs. The third layer assembles motifs into large combinations that form a close relationship with known parts of objects and later all are combined. The deep learning models (Pamina and Raja, 2019) should at least contain three layers . Every layer in deep learning receives data from the previous layer and sends it to the next layer. The deep learning models works very well with huge volume of data than the lesser data. Image classification provides a great opportunity to extend research in the field of agriculture.

Figure 1.

AI and Deep Learning

Figure 2 shows the function of deep learning. The model is trained to identify the input. The example illustrated in figure 2 is a model of identifying a Car. The network contains number of hidden layers.

Figure 2.

Deep Learning function