Accelerated Convolutional Neural Network For Iot Based Smart Farms: A Review

Introduction

Today agriculture forms the main economic activity for many countries in theworld. With the ever increasing demand for food and harsh environmental con-ditions farmers are now forced to change the production methods. As the globalpopulation is growing continuousl, a large increase of food production mustbe achieved FAO. Coping with these demands and challenges, farmers areforced to practice smart farming, which tackles the challenges of pro-duction, environmental impact and food security and sustainability.

Smartfarming is an agricultural practice that focuses on the use of information andcommunication technology(ICT); internet of things and cloud computing in thecyber-physical farm management cycle. Emerging ICT technologies relevant for understanding agricultural ecosys-tems include remote sensing, the Internet of Things IoT, cloud computing and big data analysis. Remote sensing, by meansof satellites, planes and unmanned aerial vehicles (UAV, i. e. drones) provideslarge-scale snapshots of the agricultural environment. It has several advantageswhen applied to agriculture, being a well-known, non-destructive method to col-lect information about earth features. Remote-sensing data may be obtainedsystematically over very large geographical areas, including zones inaccessible tohuman exploration.

The IoT uses advanced sensor technology to measure vari-ous parameters in the field, while cloud computing is used for collection, storage,pre-processing and modelling of huge amounts of data coming from various, het-erogeneous sources. Finally, big data analysis is used in combination with cloudcomputing for real-time, large-scale analysis of data stored in the cloud. These four technologies (remote sensing, IoT, cloud computing and big dataanalysis) could create novel applications and services that could improve agri-cultural productivity and increase food security also called smart farming [22]2 Gikunda and Jouandeauas shown in Fig. 1. for instance by better understanding climatic conditionsand changes.

This would be achieved by monitoring, measuring and analysingvarious physical aspects and phenomena continuously. The deployment of newinformation and communication technologies (ICT) for small-scale crop/farmmanagement and larger scale ecosystem observation will facilitate this task, en-hancing management and decision-/policy-making by context, situation and lo-cation awareness. This use of ICT has led to a phenomena of Big Data, that ismassive volumes of multidimensional data that can be captured, analysed andused in decision making [5]. The high volume data produced by the IoT has poseda new challenge of real-time practical processing for decision making.

One of the sucessfull data processing tool applied in this kind of large datasetis the biologically inspired convolutional neural networks (CNNs), which haveachieved state-of-the-art results in computer vision and data mining. The current survey investigates the problems that employ PCNN asubset method of Deep Learning(DL), defined as deep, feed-forward ArtificialNeural Network(ANN). Convolutional neural networks extend classical ANN byadding more ‘depth’ into the network, as well as various convolutions that allowdata representation in a hierarchical way. The motivation for carryingout the study is from the fact that CNN has been employed successfully indifferent sectors with remarkable results. However state-of-the-art networksare extremely computer memory intensive which makes them very unsuitableOptimizing IoT with PCNN 3for IoT nodes. Jayaraman et al. [? ], notes that in smart farming thousandof sensor nodes produce millions of sensor streams which is the biggest challengebecause of lack of rapid analysis of data in real time. Although model accelerationand optimization for CNN have been explored since it was brought up, recently ithas brought up huge good agricultural impact.

Therefore, in this study we wishto examine the application of PCNN models in agriculture in a move to bundleall these models in one document to help future researchers in fully optimizingagricultural CNN based tools for optimal performance and benefit to farmers. There are two ways to parallelise an IoT based cnn, across; i) modeldimension by workers training different parts of the model and ii)data dimensionby different workers train on different data examples. The ultimate objective ofparallel training.

Methodology

In the current study we shall endeavour to investigate the pcnn models that arebased on either or both data parallelism or model parallelism. The bibliographicanalysis involved two steps: (a) collection of related work; and (b) detailed reviewand analysis of these collected works. In the first step, a keyword-based searchfor conference papers and articles will be performed between Jan and September2018. The following keywords will be used in the search query: [‘acceleration’ —‘parallel convolutional neural networks’] AND [‘smartfarm’]. In this way, papersreferring to PCNN but not applied to the agricultural domain will be filteredout.

Convolutional Neural Network

CNN is a class of deep learning feed-forward ANN that has been applied sucess-fully in many computer vision and classification applications. Therehas been extensive use of cnn proposed by Lecum which is 7-layer LeNet-5structure excluding the input layer.