RSS | Register/註冊 | Log in/登入
Site search:
facebook分享

TRUE SMART AGRICULTURE

                         

Seishi Ninomiya

International Field Phenomics Research Laboratory,

Graduate School of Agriculture and Life Sciences, the University of Tokyo

 

E-mail: snino@isas.a.u-tokyo.ac.jp

 

ABSTRACT

 

Since the end of the 20th century, information and communication technology (ICT) has been dramatically advancing and changing our lives, society and industries in many aspects. Though agriculture is also highly expected to be benefitted by ICT, the application of ICT in agriculture has been rather slow compared with other industries. It is mainly because of the high complexity of agriculture which is usually conducted under uncontrolled environmental conditions with several uncertainties.

Recently, agriculture which fully harnesses ICT is often called “smart agriculture” or “smart farming”.  They sound very comfortable but we need to carefully consider what “smart” means. Simply to say, agriculture with ICT cannot be always smart if we do not understand the smartness of agriculture. Agriculture in the 20th century was rather successful in terms of producing sufficient food. The success, however, extremely depended on unlimited use of chemicals, water, energy and land, causing serious impacts on environments and biodiversity. In the 21st century, still growing population and diet transition to more animal products caused by economic growth, require continuously increasing food productivity but we cannot take the same way as the 20th century. Namely, we need to concurrently achieve both high productivity and sustainable food production. It means that we need to find the most optimal solution among extremely complicated conditions such as productivity, food safety, food quality, environmental impact, biodiversity, water use, land use, energy use, GHG emission, climatic change and farmers’ benefits.  I believe that only ICT can find such as solution and make agriculture truly “smart” even for small scale farmers. Moreover, ICT is expected to be comparatively less expensive per benefit than any other technologies and easier for famers to adopt. A very good example is the mobile phone technology.

Because of the complexity of the problem, it is not easy to solve it in a reductive approach and an inductive approach seems to be only the practical way. Such an approach in ICT is now called datacentric science. Datacentric science is a way to develop solution models such as decision support models based on observed data, sometimes leaving the models as black-boxes.

The approach needs four steps; data acquisition, data management, data analytics and modeling which are currently represented by IOT, cloud, bigdata management, artificial intelligence etc. This paper discusses about technologies and their applications for smart agriculture, covering field sensors, sensor network, crowd sensing, agricultural cloud, image analysis, machine learning, artificial intelligence, decision support models,  communication tools etc.

 

Keywords: Artificial intelligence, Bigdata, Datacentric Science, Sustainable agriculture and food production,

 

INTRODUCTION

FAO (2012) predicted that food demand would be increasing toward 2050 and that the increase would be caused not only by population growth but also by increase of per-capita food consumption, indicating the necessity of drastic improvement of food production. Agriculture in the 20th century was rather successful in terms of doubling food production. The success, however, extremely depended on unconstrained use of chemicals, water, energy and land, causing serious impacts on resource consumption, environments and biodiversity. In the 21st century, the still growing population and diet transition to more animal products caused by economic growth, require continuously increasing food productivity as FAO pointed out but we cannot take the same way as the 20th century. Namely, we need to concurrently achieve both high productivity and sustainable food production. It means that we need to find the most optimal solution among extremely complicated constraints such as productivity, food safety, food quality, environmental impact, biodiversity, water use, land use, energy use, GHG emission, climatic change and farmers’ benefits (Fig. 1). Obviously, such many constraints are making the problem so complicated. In addition, agriculture is conducted under uncontrolled environmental conditions with several uncertainties, making the solution more difficult.

Since the end of the 20th century, information and communication technology (ICT) has been dramatically advancing and changing our lives, society and industries in many aspects. Though agriculture is also highly expected to be benefitted by ICT, the application of ICT in agriculture has been rather slow compared with other industries. It is mainly because of such high complexity and uncertainty in agriculture.

 

Fig. 1. Food production in the 21st century with complex constrains.

 

Recently, new science named “data-intensive science” or “data-centric science” was advocated as the fourth paradigm shift of science after experimental science, theoretical science and computational science (Hey et. al., 2009), leading us to the boom of big data. Data-centric science is based on data, taking an inductive approach for finding out solutions rather than a reductive approach, which sometimes leave solution models black-boxes. It is said that data-centric approach is more powerful to provide solutions for complex problems as it does not require clear theories between causes and results and we are now expecting that the approach is helpful for the complexity and uncertainty in agricultural productions.

Data-centric approach generally needs the following steps; data acquisition, data management, data analytics and modeling which are currently represented by IOT, cloud, bigdata management, artificial intelligence, etc. Finally, the knowledge obtained must be properly  transferred to those who need it. This paper discusses technologies and their applications for smart agriculture, covering field sensors, sensor network, crowd sensing, agricultural cloud, image analysis, machine learning, artificial intelligence, decision support models, robotics, communication tools etc.

Recently, agriculture which fully harnesses ICT is often called “smart agriculture” or “smart farming”.  They sound very comfortable but we need to carefully consider what “smart” means. Simply to say, agriculture with ICT cannot be always smart if we do not understand the smartness of agriculture. I define smart agriculture to be one that solves or tries to solve the complex problems which agriculture of the 21st century is facing, realizing its sustainability and contributing to SDGs      (http://www.un.org/sustainabledevelopment/sustainable-development-goals/).

EXAMPLES OF ICT USE FOR SMART AGRICULTURE

Prediction of Airborne pest immigration to reduce pesticide use

Pesticides are powerful to mitigate crop damages to maintain productivity and quality. But, if we can reduce its usage, it should be ideal in terms of reducing environmental impacts, farming cost, farming labor and GHG emission from its production and transportation. Its application of optimal timing by predicting the occurrence of pests, is one of effective ways to reduce the usage amount. There have been several pest prediction models which have been proposed. Otuka et. al. (2005) proposed a dynamic model to predict immigration of an airborne insect, rice hopper to Japan from the outside of Japan. Rice hopper causes serious damages on rice (e.g. shimaha blight, Fig. 2) and farmers usually tend to apply pesticides without knowing when the hoppers immigrate into Japan, which increases the application amount. The model predicts the immigration timing and landing points in Japan. The model combines weather forecast to predict the speed and direction of low altitude jet stream which carries insects, and a particle diffusion model to predict insect dispersion with the locations of insect taking-off estimated by inverse simulation based on insect trap data (Fig. 2). By the prediction, farmers are expected to know the most optimal timing and location of pesticide applications, resulting in the usage reduction. Currently, the proposed prediction model is operational as a service for extension people as well as farmers.

Fig. 2. Prediction model for rice plant hopper immigration to west Japan (Otsuka etl al. 2005). Paddy damaged by shimaha blight (left), immigration route (middle) and immigration prediction by model (right). Figures were provided by Dr. A. Otuka.

 

YMC model to realize sustainable agriculture

The YMC (Youth Mediated Communication) model originally proposed by NPO PANGAEA (www.pangaean.org/) is an idea that school youths who are literate, bridge their illiterate parents and experts to transfer knowledge. The model was applied to a Vietnamese small village (Togami et. al., 2012) where farmers were using too much fertilizers, pesticides and too many seeds for rice production because of the lack of proper knowledge. Under the model, school youths of 9 to 14 years old periodically interview their parent illiterate farmers about their issues in farming. Then, the youths go to a village center where the Internet connected PCs are available in order to transfer the questions from the farmers to remote experts and to receive the advices from the remote experts using the PCs. The youths transfer the advices to their parent farmers (Fig. 3). The youths also work as field sensors by recording crop height and leaf color and taking pictures of unusual matters with mobile phone cameras in their parent paddy fields. The dates collected by the youths are transferred to the remote experts for them to provide proper advices to each farmer (Fig. 4). This approach was a good and successful combination of digital technology and human skill and a practical way to solve the last one-mile issue of knowledge transfer in rural area.

 

Fig. 3. YMC model for information and knowledge transfer between illiterate famers and experts via school youths. Its concept (left), interaction between youths and experts by the Internet (middle) and a tool kit for youths (right) (NPO PANGAEA http://www.pangaean.org/web/english/general/generaltop_en.html and Togami et. el. 2012).

 

Fig. 4. Data acquisition by youths in the YMC model (left), examples of the photos taken by youths (middle) and a graph of rice height by youths (right) (NPO PANGAEA, http://www.pangaean.org/web/english/general/generaltop_en.html and Togami et. el. 2012).

 

GIS-based model for optimal management of distributed small plots

One of the ordinal ways to achieve production efficiency is to expand farming land size, expecting its scale advantage and many Japanese farmers are trying to accumulate lands by renting lands from old farmers who have retired. However, what is happening in fact is an accumulation of small plots (Fig. 5, left) and very different from real large-scale farming where there are just a few large plots. To realize efficient production under such a constrained condition, ICT is highly expected to contribute for optimal precision crop management for each small plot, optimal labor and machinery arrangement to achieve productivity, crop quality and lower cost. A well-designed GIS based farm management system, PMS (Yoshida et. al., 2009) is a powerful tool for distributed small plots. PMS is now available to the public for free.

Fig. 5. A typical example of distributed plots (left) and a GIS application for the management of such distributed small plots (Yoshida et al., 2009). Figures are provided by Dr. T. Yoshida (NARO).
 

CURRENT TECHNOLOGY OVERVIEW FOR SMART AGRICULTURE

Acceleration of data acquisition IOT and multilayer monitoring

In data-centric approaches, data acquisition to provide big data is the initial point. The data acquisition must be temporally and spatially as dense as possible and multilayered combining several types of sensors to compose a sensor network and to generate big data (Fig. 6), expecting that such data will help to find new knowledge for smart agriculture.  Such a construction of a sensor network is now becoming easier as low-cost sensor nodes for agriculture are now coming to the consumer market as introduced in the next section.  

Examples of field sensor nodes

In general, three types of information are generated in the agricultural field; environmental information such as weather and soil condition, crop information such as growth status and genotypes, and management information such as fertilizer application.  Development of low-cost field environment sensor nodes started at the end of 1990 and now several companies have released commercial products. Open Field Server (Hirafuji et. al., 2010, Hirafuji et. al. 2013, Fig. 7) is a typical field sensor node which can autonomously acquire environmental data such as air temperature, soil temperature, soil moisture and solar radiation, and transmit the data to cloud servers, without any power supply. It can optionally carry a camera to watch field. Field Server has a long history of its development and currently opens its technologies to the public so that people are able to assemble the nodes cheaply and to customize it. Fig. 8 shows an autonomous three-layered soil moisture sensor (Kojima et. al., 2016) commercially available from SenSprout Ltd. (http://sensprout.com/en/). The sensor can maintain automatic data acquisition and  transmission to cloud for more than one year without any battery change. Fig. 9 shows a prototype of glove-type multispectral sensor to know fruit quality non-destructively. We expect the sensor to be used to identify most optimal timing of harvest or to phenotype in breeding when they cannot harvest until seeds are fully matured. Compared with environmental sensors, the development of crop status sensors is delayed.

 

Fig. 6. Multilayer monitoring of crop field.

Fig. 7. Open Field Server. After Hirafuji et. al. (2013)

 

Fig. 8. Fully autonomous three-layered soil moisture sensor (SenSprout).

 


Fig. 9. A glove type NIR sensor to estimate fruit quality. By Mr. Aoki of DUNAMIST, Hamamatsu Photonics Co. Ltd.

 

Sensor platform

In addition to fixed sensor nodes, mobile sensor nodes are also expected. For example, some commercial harvesters are equipped with a yield sensor so that farmers can know the yield of each small plot for better management. Agricultural machineries can be a good platform of them. Currently, the most promising mobile sensor platforms are drones. Recent advancement of drones is dynamic. Usually, handling is very easy with an autopilot and fail-safe system and the price is becoming very reasonable. Main sensors for drones are image sensors such as RGB camera, multispectral camera, hyperspectral camera and infrared camera (Fig. 10) which are expected to estimate crop canopy structure, canopy coverage, nitrogen contents, canopy temperature, grain moisture, lodging, heading timing number of panicles etc. The performance of data collection is incomparably faster that current evaluations of those characteristics done by human hands. Particularly, their spatial coverage is tremendously high. Of course, drones can be agricultural sprayer for pesticides and fertilizers as well once they secure proper payload.

Importance of human interaction

Though several autonomous sensors are available now, combination of human capability and ICT is powerful and efficient. The YMC model discussed above is one of the good examples of such combination. SNS is often discussed as a good source of big data. The approach of data collection is named crowd sourcing and it is also expected tobe useful in agricultural data collection. For example, occurrence of diseases may be able to monitor through SNS. Fig. 11 shows another example of human-IT interaction used for crop status monitoring.

 

Fig. 10. Drone and the sensors for drones. Multispectral camera, infrared camera and multispectral camera.

 

Fig. 11. Comparatively high-throughput measurement of crop status by barcode scales. Photos were kindly provided by Mr. Arturo Garcia, U. Missouri Colombia.

 

Utilization of legacy data


Though we understand the importance of big data collection, there is a serious limitation in agriculture that we can grow crops only once or twice annually in general. This means that we can have only one or two replicates at the same field and we need to wait for several years until sufficient number of data become available for a big data analysis.  An only solution is utilization of legacy or historical data recorded past. Fortunately, such kinds of data have been well preserved in many of agricultural experimental stations (Fig. 12) though many of them are not in digital formats. Currently, a few projects are being conducted in Japan to revive such data and find out how they are useful.

Fig. 12.  An example of legacy crop data stored in an agricultural experimental station.

 

Fig. 13. 3D reconstructed crop canopy by SFM/MVS.

 

3D reconstruction of crop canopy by drone images

One of the most exciting  technologies for images taken by drones is SFM/MVS (Structure From Motion/Multi View Stereo) which enables the images to reconstruct 3D structure of targets. ­­By the technology, 3D structure of crop canopy can be reconstructed from multiple images taken by a single drone (Fig. 13), making high-throughput estimation of crop height, for example.  

High-throughput crop characterization

By using images, several technologies to characterize crop phenotype have been proposed. Guo et. al. (2013) proposed an illumination invariant crop segmentation method, DTSM which can accurately separate background and crop even under varying light condition of outside field (Fig. 14) without any human interaction. It can be a powerful tool to estimate canopy coverage (Fig. 15) which is a good indicator of crop growth with correlation to LAI.  Duan et. al (2016) extended the idea to estimate canopy coverage from 3D reconstructed canopy data. Guo et. al. also proposed a method based on machine learning which can detect flowering of rice panicles (Fig. 16). With a similar approach, detection of heading of sorghum is now becoming possible (Fig. 18).

Fig. 14. Illumination invariant crop segmentation by machine learning (Guo et. al. (2013). Crop is accurately segmented regardless of  light conditions.

 

 

Fig. 15. Rice canopy coverage estimation by DTSM. R2 is as high as 0.99.

 

Fig. 16. Flowering rice panicles (left) are detected by machine (Guo et. al. 2015).

 

 

Fig. 17. Diurnal and daily pattern of rice flowering automatically detected from time series images (Guo et al., 2015). Blu and black dots indicate the number of flowering panicles counted by human eyes and machine visions respectively.

 

  

Fig. 18. Detection of sorghum heading from drone images.

 

Data interoperability

Agricultural decisions often require multiple data resources such as meteorological data, soil data and crop data which are served by different organizations. Even among the same kind of data sets such as weather databases are often maintained by many different organizations.  It is, however, very difficult to flexibly integrate those databases or to access those with a single API because of the heterogeneity of the accessibility. In fact, the difficulty has been making several agricultural systems enclosed within each of the systems and the linkage among the systems by data exchanges impossible (Fig. 19).  In order to maximally utilize distributed data as big data in total without any duplication of data generation and maintenance, a platform to make all the data interoperable is inevitable. To solve the issue, Laurenson et. al. (2002) developed a system, MetBroker to provide applications consistent accesses to heterogeneous weather databases so that we do not need to redesign programs to read data from different databases, and proved that the idea is highly useful and powerful for data sharing. Currently, several projects for agricultural data interoperability are carried out in EU, Japan, US etc.

 


Fig. 19. Currently, data exchange among different systems is rather difficult. Figure was originally provided by Dr. T. Yoshida (NARO).

 

CONCLUSION

In this article, I first emphasized that we need to understand the meaning of “Smart Agriculture” and that ICT use in agriculture is not a necessary and sufficient condition to achieve real smart agriculture though it is inevitable to solve very complex issue to make agriculture really smart.  Then, some of the good examples for smart agriculture were introduced. A part of the new technologies of ICT for agriculture was also discussed, expecting those to contribute to smart agriculture. I mainly focused on data acquisition technologies and some basic methods to extract information from acquired data which are the basis for big data construction. In addition to those technologies, we need data analytics to extract useful information from the big data to provide knowledge for decision supports to stakeholders in agriculture. In this regard, AI such as machine learning (e.g. deep learning) seems to be very powerful and promising once we have sufficiently big data and training data sets for AI. Prediction models are also highly expected by using big data. For example, it is becoming possible for us to predict crop performance (e.g., productivity) by using genotypic data and environment data. This means that we can discover the most optimal cultivar for a location without practically growing it. Agricultural robotics and plant factory are also promising fields with ICT though I did not discuss any in this article. Communication technologies are keys for knowledge and information transfer in agriculture. Particularly, they are very important to solve the last one-mile issue of such transfer to the farmers in developing countries. ICT is expected to be comparatively less expensive per benefit than any other technologies and easier for farmers to adopt. A very good example is the mobile phone technology.

REFERENCES

Duan, T., B. Zheng, W. Guo, S. Ninomiya, Y. Guo, S. C. Chapman. (2016) Comparison of ground cover estimates from experiment plots in cotton, sorghum and sugarcane based on images and ortho-mosaics captured by UAV, Functional Plant Biology, 2016. http://dx.doi.org/10.1071/FP16123.

FAO (2012) Global agriculture towards 2030/2050, The 2012 Revision, http://www.fao.org/docrep/016/ap106e/ap106e.pdf

Guo, W., T. Fukatsu and S. Ninomiya (2015). Automated characterization of flowering dynamics in rice using field-acquired time-series RGB images. Plant Methods 2015, 11:7.

Guo, W., U. K. Rage and S. Ninomiya (2013) Illumination invariant segmentation of vegetation for time series wheat images based on decision tree model. Computers and Electronics in Agriculture, Computers and Electronics in Agriculture, 96:58-66.

Hey,T., S. Tansley and K. Tolle (2009) The forth paradigm, data-intensive science discovery, https://www.microsoft.com/en-us/research/publication/fourth-paradigm-data-intensive-scientific-discovery/

Hirafuji, M., T. Fukatsu, T. Kiura, H. Hu, H. Yoichi, K. Tanaka, Y. Miki, S. Ninomiya (2010) Sensor Network Architecture Based on Web and Agent for Long-term Sustainable Observation in Open Fields, Advances in Practical Multi-Agent Systems, SCI325, Springrt-Verlag, Berlin Heidelberg, 425-433.

Hirafuji, M., H. Hideo Yoichi, Y. Miki, T. Kiura, T. Fukatsu, K. Tanaka, K. Matsumoto, N. Hoshi, H. Nesumi, Y. Shibuya, A. Itoh, S. Ninomiya, J. Adinarayana, D. Sudharsan, Y. Saito, K. Kobayashi, T. Suzuki (2013) Development of an Open Field Server and Sensor Cloud System, Agricultural Information Research, 22: 60-70

Kojima, Y., R. Shigeta, N. Miyamoto, Y. Shirahama, K. Nishioka, M. Mizoguchi and Y. Kawahara (2016). Low-Cost Soil Moisture Profile Probe Using Thin-Film Capacitors and a Capacitive Touch Sensor. Sensors, 16(8), 1292.

Laurenson, M. R., T. Kiura and S. Ninomiya (2002) Providing agricultural models with mediated access to heterogeneous weather databases. Applied Engineering Agric. 18: 617-625.

Otuka, A., T. Watanabe, Y. Suzuki and M. Chino (2005) Real-time prediction system for migration of rice planthoppers Sogatella furcifera (Horvath) and Nilaparvata lugens (Stal) (Homoptera: Delphacidae), Appl. Entomol. Zool., 40:221-229.

Togami, T., S. Ninomiya, K. Yamamoto, Y. Mori, T. Takasaki, Y. Okano, R. Ikeda, A. Takezaki, T. Kameoka (2012). Field and weather monitoring with youths as sensors for agricultural decision support, Agricultural information Research 21:2012,65-75.

Yoshida, T., H. Takahashi and I. Teramoto (2009) Development of Planning and Management Support Software for Farm Work Using a Geographic Information System-Compatible Field Map, Agricultural Information Research, 18:187-198Acharya, S.N. and H.C. Huang. 2000. Registration of ‘AC Longview’ alfalfa. Crop Science 40(6):1823.


Download the PDF. of this document (101), 1,964,636 bytes (2 MB).