Author(S) Repeat Author and Affiliation Boxes As Needed

Send your completed paper to Sandy Rutter at by 13 April 2007 to be included in the ASABE Online Technical Library.

If you can't use this Word document and you'd like a PDF cover sheet please contact Sandy.

Please have Word's AutoFormat features turned OFF and do not include live hyperlinks. Your paper should be no longer than 12 pages. For general information on writing style, please see

This page is for online indexing purposes and should not be included in your printed version.

Author(s)

First Name / Middle Name / Surname / Role / Email
Mirwan / Ushada / Graduate Student, ASABE Member No: 1027757 /

Affiliation

Organization / Address / Country
Osaka Prefecture University, Graduate School of Life and Environmental Sciences, Department of Applied Life Sciences, Laboratory of Bioinstrumentation, Control and Systems (BICS) Engineering / 1-1, Gakuen-cho, Naka-ku, Sakai Japan ZIP 599-8531 / Japan

Author(s) – repeat Author and Affiliation boxes as needed--

First Name / Middle Name / Surname / Role / Email
Haruhiko / Murase / Professor, ASABE Member No: 9349 /

Affiliation

Publication Information

Pub ID / Pub Date
073104 / 2007 ASABE Annual Meeting Paper

The authors are solely responsible for the content of this technical presentation. The technical presentation does not necessarily reflect the official position of the American Society of Agricultural and Biological Engineers (ASABE), and its printing and distribution does not constitute an endorsement of views which may be expressed. Technical presentations are not subject to the formal peer review process by ASABE editorial committees; therefore, they are not to be presented as refereed publications. Citation of this work should state that it is from an ASABE meeting paper. EXAMPLE: Author's Last Name, Initials. 2007. Title of Presentation. ASABE Paper No. 07xxxx. St. Joseph, Mich.: ASABE. For information about securing permission to reprint or reproduce a technical presentation, please contact ASABE at or 269-429-0300 (2950 Niles Road, St. Joseph, MI49085-9659USA).

An ASABE Meeting Presentation

Paper Number: 073104

Development of an Intelligent Quality Control Model Based on Speaking Plant Approach and Kansei Information for Moss Greening Product

Mirwan Ushada, Graduate Student, ASABE Member

Osaka Prefecture University, Graduate School of Life and Environmental Sciences, Department of Applied Life Sciences, Laboratory of Bioinstrumentation, Control and System (BICS) Engineering, 1-1 Gakuen-cho Naka-ku, Sakai Japan ZIP 599-8531,

Haruhiko Murase, Professor, ASABE Member

Written for presentation at the

2007 ASABE Annual International Meeting

Sponsored by ASABE

MinneapolisConvention Center

Minneapolis, Minnesota

17 - 20 June 2007

Abstract.In this study, sub-systems of intelligent quality control based on speaking plant approach and kansei information were proposed.It consists of quality and quantity (growth) model.It utilizes Artificial Neural Network (ANN),plant response,kansei index and texture analysis.The first ANN model for quality is proposed to define the relationship between textural features and kansei index.Kansei index is measured using visual appearances as the representation of plant factory owner. The target point of the model is customer of moss product.The second ANN model for growth is proposed to define the relationship among plant response, textural features and temperature.Plant response is measured by using wet weight. The target point of the model is plant factory parameter.

Four cycles of re-watering treatment were done based on two different local environments inside the same optimum environmental set point (global). Temperature of 100C and RH of 75% was considered as the optimum environment for the moss. The textural features have shown the various pattern compared with the changes of wet weight. It shows the difference pattern with our previous research (Ushada et al., 2006a) due to occurrence of growth.

The research result shows that texture analysis is possible to be used as pattern recognition tool not only for quality but also for growth model. The first ANN model with satisfied inspection errorcan be used to predict the customer preferences while the second ANN modelwith satisfied inspection error can be used to predict the optimum local temperature.

Keywords. Artificial neural network, Kansei index, Local environment, Textural features, Wet weight

(The ASABE disclaimer is on a footer on this page, and will show in Print Preview or Page Layout view.)

Introduction

Recently in Japan, mass bio-production of moss plant has been expected because of the huge demand of roof top and wall greening for buildings. It has a function to ease the urban heat island effect. In order to fulfill the customer demand on quantity and quality, Murase (2004) has developed a new technology which grows the cultured moss in one to two months compared to the three years in natural growth. It has been produced from plant factory.

Increasing consumer wealth comes with increasing consumer demand for better quality agricultural products. In order to yield better quality plants for this market, advanced control techniques is needed (Hashimoto et al., 2006). Hashimoto et al. (1985) has highlighted the control system named as Speaking Plant Approach (SPA) based on physiological responses to determine the physiological status, and then use the information for environmental control for optimization.

In other side, Ujita and Murase (2006) have highlighted the importance of Profile of Mood Status (POMS)to attract the customer to buy a product from psychological aspect. Agricultural industry such as plant factory has realized the importance to understanding customer preferences and incorporates them into production system. When asked to describe preferences from greening material product, costumer will frequently include a mixture of functional features and descriptions relating to how the design appeals to them on a more subjective emotional level (for example, eye catching, free maintenance and physiological status). Ushada et al. (2006b) has highlighted the possibility to detect the water status using visual appearance as representation of plant factory owner. The methods which were performed by Ujita and Murase (2006) and Ushada et al. (2006b) can be categorized as Kansei engineering. It is a technique aimed at translating subjective requirements in to design features and incorporating preferences into the product design process.

Figure 1. On-demand production scheme using Kansei information control

According to Nagamachi (2001) there are three focal points of Kansei Engineering: 1) How to accurately understand consumer Kansei; 2) How to reflect and translate Kansei understanding into product design; 3) How to create a system and organization for Kansei orientated design. These three points were accommodated in to prototype of on-demand production scheme using kansei information control (fig. 1). This proposed system will accommodate artificial intelligence tools such as Artificial Neural Network (ANN), Bayesian belief network, pattern recognition such as texture analysis and Kansei information method. Finally it will be regulated by system controller.

Kansei information can be approached by indexing images based on the inner impression experienced by a person while viewing an image. The aspects of the image that evoke this impression are called Kansei factors. In previous research, Kobayashi and Kato (2000) have researched subjective interpretation for statistical texture images in order to clarify human subjective interpretation mechanism for images. They focused on orientation interaction and color property, and unified those features by calculating the contrast over the entire image resolution. Kawasumi et al. (1999) also has proposed a new method to analyze the factor of human feeling by using an experimental color simulator. It enables to extract the colorimetric factors concerned with the feeling of depth (FOD) which arises psychophysically when we look into a painted automobile panel. In this study, kansei information was approached using the kansei index combining visual appearances and textural features.

System controller of fig.1 can be approached by using SPA. Ushada et al. (2007) has developed the first two sub-systems of Intelligent Quality Control (IQC) for SPA covering non-destructive sensing and its inverse model using texture analysis and artificial neural network.

In this study, the second two sub-systems of IQC based on speaking plant approach and kansei information were proposed. It consists of quality and quantity (growth) model. It utilizes ANN, plant response, kansei index and texture analysis. The first ANN model for quality is proposed to define the relationship between textural features and kansei index by extracting the information from the both models for canopy parameters in Ushada et al. (2007). Kansei index is measured using visual appearances as the representation of plant factory owner. The target point of the model is customer of moss product. The second ANN model for growth is proposed to define the relationship among plant response, textural features and temperature. Plant response is measured by using wet weight. The target point of the model is plant factory parameter. The purpose of our study is to construct a prototype of on-demand production system using kansei information control. The specific objective is to identify the relationship between plant response and textural features on the different environment.

Materials and Methods

Greening material

Set samples of high density and immature colony-based type of cultured sunagoke moss (Rhacomitriumcanescens) used in roof top greening were monitored in growth chamber, Biotron NK-50. The moss was placed in plastic case anti static. The optimumgrowth chamber set point of temperature of 100C-relative humidity of 75% was identified. Two different locations inside growth chamber which has different local temperature of 140C and 15.90C were used.

Kansei indexing factors

The focus of kansei indexing methods is on the viewer, rather than on the image, and similarity measures derived from kansei indexing represent similarities in inner experience, rather than visual similarity. It is now widely acknowledged that image contain multiple levels of visual content. For example, luminance, and color are regarded as low-level content, and physical objects are regarded as high level content, while textures and patterns are often regarded as mid-level content (Black et al., 2004).

In this paper while watching images of moss canopy, the writer was assumedas a representation of plant factory to evaluate comprehensively each canopy considering 4 (four) levels (Soak, Wet, Semi-Dry and Dry) using 1 (one) visual information that is canopy appearance based on texture analysis.

The classification scale is usually consisting of discrete values of assessment such as excellent, very good, good, bad, very bad, high quality, medium, low quality and dry. As shown in Table 1, in the textural classification, the class is varied from soak, wet, semi-dry and dry status. The requirement on these statuses is based on the water status response to the environment in speaking plant approach method. The class will be one of parameters whether the environment is fit to the growth of moss. The class is indexed in to the value between 0 and 1 in order to fit the sigmoid function of the proposed ANN

Table 1 Classification on water status of moss

No / Status / Scale
1 / Soak / 1
2 / Wet / 0.75
3 / Semi-Dry / 0.25
4 / Dry / 0

Texture analysis

The co-occurrence matrix or known as grey tone spatial dependence matrix is a second order image variation. It can provide a basis for a number of textural features of image (Haralick et al., 1973; Murase et al., 1994).

The following textural features are the most critical to plant growth indices (Murase, et al., 1997):

(a)Energy:

(1)

(b)Contrast:

(2)

(c)Local Homogeneity (LH):

(3)

where :

d = distance between two neighbouring resolution cells

q= angle between two neighbouring cells

P(1,0)(i,j)= joint probability density function at d = 1 and q =0

i, j= notation for grey tone

The capability of texture analysis as a pattern recognition tool in plant factory can be shown in fig.2. It can make two different pattern of moss greening material due to cultivar.

Figure 2. Detecting different cultivar using textural features

Results and Discussion

Quality model

Inverse model of non-destructive sensing has been proposed in the other work in term of application for simple vision model as shown in fig. 3 (Ushada, et al., 2007). It is also defined as Forward Kansei Engineering Method because it can only be used to convert the consumer’s Kansei into design parameters (Schutte, 2002).

Figure 3. Forward kansei engineering model (Ushada et al., 2007)

The recent system is called Backward Kansei Engineering System because it can be used to predict the Kansei that a user will have from a drawing or concept (Schutte, 2002). This model will be proposed as kansei indexing factor using ANN (fig.4).

Figure 4. Backward kansei engineering model

ANN model consists of 3 input neurons, 5 hidden neurons and 1 output neuron. The model reaches the convergence during 20.000 iterations (fig.5). The learning error is 0.095 while the validating one is 0.084. It shows the satisfied performance of ANN model to predict the kansei index as customer parameters.

Figure 5. Learning convergencefor backward model

Growth model

Figure 6. Detecting growth based on different environment using textural features

Notes:

I = Raw materialV= Recovery I

II= Wilting IVI= Recovery II

III= Wilting IIVII= Recovery III

IV= Wilting IIIVIII= Recovery IV

The initial set point of global environment was temperature of 150C and relative humidity of 60% (Wilting). After observing several days which concluded the improper environment, it was changed in to 100C and 75% (Recovery). The obvious different effect between 2 environmental conditions to textural features can be shown in fig.6.

Figure 7. Wet weight profile due to re-watering

Based on the result, temperature of 100C and RH of 75% was considered as the best environment for the moss by evaluating its texture and appearance. The measurement of wet weight was started using the best environment. Hashimoto et al. (2006) suggested that fusing various types of information on plant response allows a correct estimation of the present physiological status of the plant.

In this study the comparison between textural features and wet weight was highlighted as control point.Figure 7 shows wet weight profile as the plant response which was affected by re-watering treatment. The time unit is defined as random sampling of experimental day. 48 data of 2 different samples on different places inside the growth chamber was collected. During the observation, 4 cycles of re-watering has been done.

As the plant enters the wet status, it has a high water content which is reflected by higher value of wet weight. Due to environmental effect, it reaches the lower value of wet weight. The textural features have shown the variouspatterns with the wet weight (Figs. 8a, 8b, 8c and 8d). This result is different with our previous research (Ushada et al., 2006a) due to different objective treatment. In the previous research, the treatment is subjected to drying condition while in this research, by looking the best environment condition based on the change from wilting stage into growth stage reflected by textural appearance. It shows that plant response from texture analysis and wet weight has shown different pattern due to growth.Texture analysis is possible to be used as pattern recognition tool not only for quality but also for growth model.

(a)

(b)

(c)