Professional Documents
Culture Documents
Smart Forming PDF
Smart Forming PDF
Introduction
With the growing adoption of the Internet of Things (IoT), connected devices have penetrated
every aspect of our life from health and fitness, home automation, automotive and logistics, to smart
cities and industrial IoT. Thus, it is only logical that IoT, connected devices, and
automation would find its application in agriculture, and as such, tremendously improve nearly
every facet of it. Farming has seen a number of technological transformations in the last decades,
becoming more industrialized and technology-driven. By using various smart agriculture gadgets,
farmers have gained better control over the process of raising livestock and growing crops, making
it more predictable and improving its efficiency.
IoT platforms that can utilise IoT devices provided independently by a third party must
provide for sensor discovery (i.e., ability to search sensor metadata). Although there has been
extensive work in providing metadata management for sensory data, no IoT platform currently fully
supports sensor discovery. For instance, projects such as SensorMap have visualised live sensor data
on a map, allowing users to find sensors based on the sensor location. SmartFarmNet takes
advantage of semantic web principles (such as Linked Open Data and Semantic Sensor Ontology
(SSN)) to search for and share sensor information in a way that can be read automatically by
computers. Improving farm productivity is essential for increasing farm profitability and meeting
the rapidly growing demand for food that is fuelled by rapid population growth across the world.
Farm productivity can be increased by understanding and forecasting crop performance in a variety
of environmental conditions. Crop recommendation is currently based on data collected in field-
based agricultural studies that capture crop performance under a variety of conditions (e.g., soil
quality and environmental conditions). However, crop performance data collection is currently slow,
as such crop studies are often undertaken in remote and distributed locations, and such data are
typically collected manually. Furthermore, the quality of manually collected crop performance data
is very low, because it does not take into account earlier conditions that have not been observed by
the human operators but is essential to filter out collected data that will lead to invalid conclusions
(e.g., solar radiation readings in the afternoon after even a short rain or overcast in the morning are
invalid, and should not be used in assessing crop performance). Emerging Internet of Things (IoT)
1
technologies, such as IoT devices (e.g., wireless sensor networks, network-connected weather
stations, cameras, and smart phones) can be used to collate vast amount of environmental and crop
performance data, ranging from time series data from sensors, to spatial data from cameras, to human
observations collected and recorded via mobile smart phone applications. Such data can then be
analysed to filter out invalid data and compute personalised crop recommendations for any specific
farm. In this paper, we present the design of SmartFarmNet, an IoT-based platform that can automate
the collection of environmental, soil, fertilisation, and irrigation data; automatically correlate such
data and filter-out invalid data from the perspective of assessing crop performance; and compute
crop forecasts and personalised crop recommendations for any particular farm. SmartFarmNet can
integrate virtually any IoT device, including commercially available sensors, cameras, weather
stations, etc., and store their data in the cloud for performance analysis and recommendations.
SmartFarmNet is the first and currently largest system in the world (in terms of the number of sensors
attached, crops assessed, and users it supports) that provides crop performance analysis and
recommendations.
To observe the growth of the crop under varying real-world conditions (e.g., soil quality,
environmental conditions, etc.), typical crop studies involve phenotyping to understand the key factors
(e.g., the pH levels of soil, the rate of Nitrogen depletion) affecting growth. Such studies are conducted
in natural outdoor environmental conditions and locations where plants are growing, by varying
irrigation and the application of fertilizers/additives. Internet of Things (IoT) technologies
can lower the cost and increase the scale of such studies via the collection of related time series data
from sensor networks, spatial data from imaging sensors, and human observations recorded via mobile
smart phone applications. For example, IoT devices can help to capture the pH levels of soils and the
rate of Nitrogen depletion as time-series data, and share it among interested researchers and growers
for further analysis. Point solutions for smart farming currently exist, but they can only utilise a small
number of specific IoT devices (e.g., a specific model of soil humidity sensor), and provide no support
for data analysis or sharing. Using such existing solutions also requires a significant effort in order to
integrate and correlate the data obtained from different IoT devices , e.g., data from a fertilizer sprayer
on a tractor (made by one manufacturer) with the data obtained from soil moisture sensors (made by
a different manufacturer). Existing solutions are not designed around a bring-your-own IoT sensor
principle that will allow the use of new IoT devices without modification, and permit such solutions
2
to keep up with the rapid development of cheaper and better IoT sensors. Furthermore, none of the
existing solutions are designed for comprehensive and scalable analysis,
recommendation/visualisation, or sharing of crop performance data among farmers, growers,
biologists, government, and commercial organisations that support farming operations and produce
relevant products.
3
2. Related Work and Challenges is Building IoT-Based Platforms for Smart
Farming
Relational database systems (e.g., mysql [2]) have been the historic choice for storing data from the
sensors for later retrieval, analysis, and visualisation. However, the performance of these systems can
be compromised by large streams of time-series data, and they do not address most of the challenges.
To meet such challenges vendors such as John Deere [3], IBM [4], and SensorCloud [5] currently
explore or provide proprietary solutions that support specific sensors. However, vendor-specific
solutions provide limited or no interoperability with other devices in the IoT ecosystem. Vendor lock-
in is also a serious issue.
A plethora of IoT middleware platforms has recently appeared in the marketplace. Examples include
UBIDOTS [6], Xively [7], Thing Speak [8] Open.Sen.se [9], and SensorCloud [5] and more recently,
Amazon IoT [10] and IBM IoT [11] platforms. These IoT middleware platforms aim to simplify the
injection of data from all kinds of sources (physical devices, human input, online data, etc.) using a
common Application Programming Interface (API). All such platforms typically provide basic
functionalities for filtering and aggregating the data, as well as specification of events based on input
data. Although the IoT platforms in this category (e.g., those based on IBM (Bluemix) and Amazon
(EC2)) provide tools for data ingestion, storage, and computation via message passing interfaces
utilising message queues and publish/subscribe mechanisms, the benefits of these IoT platforms in
smart agriculture is yet to be determined.
Existing IoT-platforms are not designed to support near-real-time
data ingestion, quick analysis, and visualisation of large volumes of sensor data. One of the key
features of the SmartFarmNet platform is in its ability to deal with the high velocity of sensor data
(time-series) by providing solutions for fast and scalable data analytics and retrieval. The do-it-
yourself principles-driven user-interface allows complex data workflows to be built, composed, and
executed without a need for programming. SmartFarmNet also provides a web-based virtual
laboratory environment for biologists, farmers, and scientists to manage large volumes of crop sensor
data while supporting rapid responses to queries, real-time user interaction, and the ability to share
data from studies performed by multiple researchers.
4
Platform Sensor Discovery Bring-Your-Own Scalable Data Sharing Sensor,
Data,
IoT Device Analysis
and Analysis Results
(Virtual Lab)
UBIDOTS Not Supported Yes, but requires No No. Only provides
considerable efforts to API for raw data access
develop new interfaces
Xively Partial support with no Yes, but requires No No. Only provides API
specific approach for considerable efforts to for raw data access
metadata develop new interfaces
description/management
SensorCloud Not supported Supports only Partial Partial
vendor-specific sensors
(some support for CSV
file
data)
IBM Bluemix Not supported Yes, but requires Partial with additional Partial by using existing
considerable efforts to development required bluemix infrastructure
develop new interfaces as a service platform
Amazon IoT Not supported Yes, but requires Partial, with Partial by using existing
considerable efforts to additional EC2 infrastructure as a
develop new interfaces development required service platform
IoTCloud Not supported Yes, but requires No No
considerable efforts to
develop new interfaces
Apache Not supported Yes, but requires No No
Storm considerable efforts to
develop new interfaces
SmartFarmNet Supported via Semantic Yes with in-built support Yes, real-time data Yes, easy to use
Web Technologies for analytics functions are e-commerce-like use
30+ commercial and built in interaction model
experimental sensors
(Table 1. Comparison of SmartFarmNet with other Internet of Things (IoT) platforms. EC2: Elastic Cloud
Computing)
5
environmental, soil, fertilisation, and irrigation conditions, including weather conditions, plant
canopy temperature, soil moisture, soil quality and salinity, fertiliser usage, and irrigation. Phenonet
is helping plant biologist and growers achieve the following:
(1) Identify the influence of different conditions on a variety of crops in real-world outdoor farm
environments;
(2) Understand water resource consumption in order to manage it effectively;
(3) Study the impact of various fertilizers;
(4) Get real-time data to forecast crop performance; and
(5) Share data and results.
6
insights with growers and farmers, as well as related scientific organisations. Some of the main
challenges in supporting such studies and also normal farming activities include the following:
Capturing the large volume of heterogeneous data produced by a variety of IoT sensors (and
possibly manual measurements), and doing this for a large number and variety of activities involving
different studies as well as crops. Supporting the integration and use of almost any IoT device,
including all commercially available sensors, camera, weather stations, etc., as this will achieve a
bring-your-own sensor model of operations that will allow farmers, growers, and scientists to take
advantage of cheaper/more capable IoT sensors, as well as individual preferences and budgets.
Integrating heterogeneous data from such a great variety of IoT devices and also historical crop
performance data produced by past studies (such data and results are typically available in CSV files
that make it harder to use, analyse, explore, and share). Providing crop performance data analysis
software and related tools for do-it-yourself search, analysis and visualisation of collected data across
multiple studies.
4. The SmartFarmNet IoT Platform for Smart Farming
SmartFarmNet is the first and currently largest system in the world (in terms of the number of sensors
attached, crops assessed, and users it supports) that provides crop performance analysis and
recommendations.
4.1 SmartFarmNet’s Data Model
User is a logical entity representing a project, a research group, or any other entity that owns data
from a study. Experiments are crop studies, and each experiment has only one owner user. Tests are
comprised of a collection of nodes, but each node can belong to multiple crop studies. A node is a
group of data stream. A stream is a time series data from an experiment with a unit of measurement.
Metadata can be attached to every hierarchical layer of the data model.
The layout of a typical crop study (referred to as “experiment” in the data model) is illustrated in
Figure 2. Such studies involve a particular area (one or more plots) in a field. Any farm can be divided
into plots, and the granularity of the analysis supported by SmartFarmNet is currently 9–10 square
meters. The plots are mapped to the node level that represents the sensor network platform (hosting
all related communication and processing capabilities). Physical sensors observing a physical
phenomenon (e.g., gypsum block sensors measuring soil moisture at multiple depths) are connected
to corresponding streams. A measurement of soil moisture at a particular depth is mapped to a data
stream associated with a node. In summary, the data streams map to the physical or virtual sensors
7
that monitor a phenomenon, while nodes and experiments/studies are used for grouping streams at
different levels. The metadata associated with nodes and streams at each level are critical for providing
contextual information. In Phenonet, metadata at the experiment/study level may include the year
when the study was conducted, the date the crop was sown, a description of the objectives of the
study, and even descriptions about study site (e.g., the soil type). At the node level, the most important
metadata fields are the plants’ genotype and its relative location within the plot (in most cases, a
row/column notation is used). Treatments applied to individual plants can also be appended as
metadata at the node level. The most relevant metadata at the stream level for Phenonet are the depth,
the sensor type, and the sensor serial number, while sensor information such as the date of calibration
or setting of the sensor can also be critical.
8
SSN concepts across multiple sensor platforms. The four main principles of linked data include: (1)
using unique Uniform Resource Identifier (URIs) to represent each IoT thing; (2) providing HTTP
interfaces to access the URIs (descriptions); (3) offering information related to URIs via the Resource
Description Framework (RDF) ; and (4) linking URIs. The SmartFarmNet ontology is an extended
version of the SSN ontology that describes the SmartFarmNet data model. The SmartFarmNet
ontology is an extended version of the SSN ontology that describes the SmartFarmNet data model.
The ontology provides the means to map the data model to sensor descriptions. For example, a soil
moisture sensor X is deployed in a plot Y, which grows a crop Z. Sensor X and its data are described
using the SSN ontology. This ontology includes the physical and technical nature of the installed
sensor network, such as what physical platform the sensor is on, and location of the sensor. The
description of the plot Y and crop Z is based on the Descriptive Ontology for Linguistic and Cognitive
Engineering (DOLCE) Ultra-Light upper ontology. The SmartFarmNet Ontology describes the plots,
the crops and the corresponding treatments applied to the crops. For example, consider crops with a
particular genotype (e.g., revenue) that are sown in plots belonging to a study. This crop is subjected
to different treatment (e.g., irrigation). The sensors deployed in the plots are governed by a geo-
location observation system either at the plot level, or at the level of some sub-feature of the plot (e.g.,
a layer of soil within the plot), or as the entire site of the study (e.g., temperature and humidity
covering the whole study site). The sampling feature links the sensors to corresponding plots, and
from the plot to the combination of genotype, treatments, and events (such as sowing dates, etc.).
Features can be related such as a soil layer can be from an individual plot or from several plots linked
together into a study site.
4.3 Scalable Data Analysis
Sensor Data in SmartFarmNet: SmartFarmNet provides near real-time response to queries over time-
series data streams from sensors. Just like in the SensorDB [4] platform we developed earlier, the
SmartFarmNet platform uses an internal data structure called micro summarisation to build a
summary of statistical features for incoming sensor data streams. SmartFarmNet uses a combination
of non SQL (NoSQL) and Semantic data stores to manage user data, sensor data, aggregated data, and
caching of commonly used information. The storage layer behind each aggregation window is
allocated based on query access patterns. Using this approach, we can optimize micro-summary
computations for different scenario.
9
The data stream processing model of SmartFarmNet is presented in Figure 4. The platform
provides three levels of data storage: (1) Short Term Cache; (2) Moving Summaries; and (3) Static
Summaries. The summaries computed over the moving aggregate window are termed “moving
summaries”, and are updated with new incoming data streams. The static summaries are provided for
data streams whose data does not change rapidly over time, hence not requiring frequent re-
computation of summaries. Finally, the short-term cache is used to store raw data in order to improve
the response time of raw data requests.
SmartFarmNet uses a combination of non SQL (NoSQL) and Semantic data stores to
manage user data, sensor data, aggregated data, and caching of commonly used information. The
storage layer behind each aggregation window is allocated based on query access patterns. Using this
approach, we can optimize micro-summary computations for different scenarios. For instance, if a
data item at the 1-h aggregation window is accessed more frequently than the 1-min aggregation
window, SmartFarmNet can be configured to store and cache data in that collection window in faster
storage, such as in memory storage or a solid state disk drive. The data stream processing model of
SmartFarmNet is presented in Figure 4. The platform provides three levels of data storage: (1) Short
Term Cache; (2) Moving Summaries; and (3) Static Summaries. The summaries computed over the
moving aggregate window are termed “moving summaries”, and are updated with new incoming
data streams. The static summaries are provided for data streams whose data does not change rapidly
over time, hence not requiring frequent re-computation of summaries. Finally, the short-term cache
is used to store raw data in order to improve the response time of raw data requests. The resource
allocation strategy uses the combination of the query and the performance of each storage to determine
where to store incoming data streams. The unique feature of the platform is its elasticity. A stateless
process thread works on the queue (at the left of Figure 4) to process each individual sensor data
stream. Therefore, we can achieve a high degree of distributed and parallel processing that can utilise
available computing resources very efficiently. The stream maintains the stream identifiers that link
it to the moving micro-summaries of all related streams computed over days, weeks, months, and
years.
10
(Fig. 4 SmartFarmNet—scalable data analysis of sensor data[1])
11
(Figure 5. SmartFarmNet architecture. DIY: do-it-yourself; RDF: resource description framework;
API: Application programming interface.[1])
12
4.4.2 Cloud Store
This enables the storage and management of data streams generated by the SmartFarmNet gateway.
The cloud store uses the linked sensor middleware-light (LSM-Light) component of OpenIoT [31].
The cloud infrastructure stores all of the relevant sensor annotations (descriptions and metadata), the
ontology, functional data related to user accounts, and permissions to enforce privacy and security.
SmartFarmNet uses a semantic data store (a No-SQL graph database) to store the sensor data in RDF
format. It internally implements publish/subscribe queues to handle large streams of sensor data
stemming from virtual and physical sensors. The components provide APIs to perform the basic
create, read, update and delete (CRUD) operations over the sensor data, and is responsible for
transforming the data generated by the IoT devices into RDF triples using the SmartFarmNet
ontologies.
4.4.3 Sensor Explorer:
This responsible for sensor discovery function, a novel feature of the SmartFarmNet platform. The
detection feature uses ontology matching, allowing data collected to be re-used/re-purposed (i.e., re-
purpose data gathered from one study for another study). For example, using the SmartFarmNet
ontology, a semantic reasoner can discover data sources (IoT devices) that measure canopy
temperature data deployed in plots that have had a specific nitrogen fertilizer treatment applied (study
property). The discovery process provides much more features than a simple filtering operation by
using semantic matching and reasoning methods. The sensor explorer uses the Scheduler and Service
Delivery and Utility Manager of OpenIoT for sensor and data discovery.
As mentioned before, the SmartFarmNet platform is built on the widely used open source platform
for Internet of Things, namely OpenIoT, available for free download. Table 2 presents the
implementation details of the platform. The entire platform is deployed in a JBOSS application
container, while the support for real-time scalable data analytics is built independently using Redis,
an in-memory data store.
13
COMPONENT IMPLEMENTATION DETAILS
JAVA-based semantic sensor stream
processor. Arduino and ArduCrop sensor
SMARTFARMNET GATEWAY (X-GSN)
wrappers to interface with IoT devices
CLOUD DATA STORE (LSM-LIGHT) LSM-Light developed using JAVA and Open
Virtuoso triple store.
SENSOR EXPLORER Java applications deployed in JBOSS
REASONER SERVICE Apache Jena supported by SmartFarmNet
OWL ontology
USER INTERFACES Do-it-yourself tools developed in Java Server
Faces (JSF)
DATA ANALYTICS Redis
Table 2. SmartFarmNet platform—implementation details. LSM-Light: linked sensor middleware-light.
SmartFarmNet is designed with the Bring-Your-Own sensor principle and supports virtually any
IoT provided by any vendor. The Sensor Schema Editor (SSE) enables users to describe and register
a sensor with ease and consistency. This process hides the complexities of editing OWL files (used
to represent ontologies) from the users. The Sensor Schema Editor provides annotations to sensors
and sensor-related data, using the underlying ontology and Linked Data principles. The interface
automates the generation of RDF descriptions for sensor node information submitted by the users.
The SmartFarmNet platform user interface (UI) allows end-users to explore and analyse agricultural
study data with zero-programming efforts. The UI hides the complexities of semantics and ontological
representations from the user by presenting concepts that the user is familiar with and understands
(e.g., plots, studies, sensor device, barcode, etc.), as defined in the data model and the ontology.
14
5.1.2.1 Resource Discovery
The user interface allows a user to search for sensors based on domain-based criteria, including
genotype, crop treatment, and the barcodes used to identify studies, as well as sensor-based criteria,
such as specific platform or location dynamically obtained from the ontology.
The provide search interface presents a list of sensors matching the discovery criteria provided by a
user. The user can then compose a query targeted to the selected sensors. Internally, this makes a
request to the Sensor Explorer and the real-time statistical analysis components.
5.1.2.3 Service Visualisation
This visualizes the fetched data using a one of the provided visualisation outputs, such as the time-
series graph.
6. SmartFarmNet Evaluation
An essential design requirement of the SmartFarmNet platform is to be able to scale, store, process,
and several hundreds and thousands of IoT sensors. The volume of data generated by the IoT sensor
is not an issue. However, the velocity at which the data is produced is very high and results in billions
of sensor data points. The objective of this evaluation is to evaluate the scalability of the
SmartFarmNet platform. In particular, we evaluate the proposed real-time statistical analysis feature
of the SmartFarmNet platform used to deliver sub-second query response latency while handling high-
velocity data streams.
Conclusions
This presents SmartFarmNet, a pioneering effort in building a scalable sensor data acquisition,
analysis, and visualisation platform for smart farming applications, based on the Internet of Things.
We presented the architectural design of the platform that aims to support virtually any IoT devices,
allow rapid ingestion and visualisation of IoT data using zero programming effort (do-it-yourself
principles), and provide a virtual laboratory environment for visualisation and sharing of study data.
The proposed SmartFarmNet uses a unique and novel real-time statistical analysis approach that
enables near real-time responses to user queries (validating the platform’s ability to scale in order to
handle high-velocity data streams). Through evaluation using actual farming data, we validated the
elasticity and scalability of the platform.
15
References
1. Jayaraman, P., Yavari, A., Georgakopoulos, D., Morshed, A., & Zaslavsky, A. (2016). Internet
of things platform for smart farming: Experiences and lessons learnt. Sensors, 16(11), 1884.
2. MySQL, 2016. Available online: https://www.mysql.com (accessed on 10 July 2016).
3. Big Data-Startup, 2016. Available online: https://datafloq.com/read/j (accessed on 6 July
2016).
4. Analytics in Agriculture, 2016. Available online: http://www-01.ibm.com/ (accessed on 6 July
2016).
5. SensorCloud, 2016. Available online: http://www.sensorcloud.com (accessed on 10 July
2016).
6. UBIDOTS, 2016. Available online: http://www.ubidots.com (accessed on 10 July 2016).
7. Xively, 2016. Available online: http://xively.com (accessed on 10 July 2016).
8. Think Speak, 2016. Available online: http://www.thinkspeak.comcom (accessed on 10 July
2016).
9. opensen, 2016. Available online: http://open.sen.se (accessed on 10 July 2016).
10. Amazon IoT Platform, 2016. Available online: https://aws.amazon.com/iot/how-it-works/
(accessed on 3 July 2016).
11.IBM Watson Internet of Things, 2016. Available online: http://www.ibm.com/internet-of-
things/ (accessed on 18 July 2016).
16