1. Image Recognition System in Fields National Agriculture and Food Research Organization and University of Tsukuba Kei Tanaka

2. Contents Backgrounds – Field Server – Features of Image Image Change Detection Application – Server – Web Application Results Conclusions

3. Backgrounds (1) A Field Server is a Web server that has a weather station and a camera. Field Server has been installed in many fields all over the world. A standard Field Server takes an image every 2 minutes. A total of 720 images for each Field Server are stored every day.

4. Backgrounds (2) Existing application Image viewer shows images as thumbnail or animation. (Web application & Java Applet) More effective use of images is expected. observing the growth of plants detecting the changes in a field Animation Thumbnail

5. Features of Image Intermittent image. Changes do not occur fundamentally between the images taken in the field. Users want to see only the images in which change took. – appearance / disappearance of people or cars – density change by daylight – shaking of a tree by wind – clouds flowing past These unimportant changes should not be detected.

6. The image with appearance of this day is only this one.

7. Image Change Detection System Image Change Detection Server Image Change Detection Program Record image change information on XML format files Change Information Web application Thumbnail Timeline Field Server Stored Images Images

8. Change Detection Algorithm 1.Calculate the joint histogram, which is a two-dimensional histogram of combinatorial intensity levels, (I 1 (x), I 2 (x)), where I k represents intensity levels (0 – 255) of each image. 2.Select clusters expected to correspond to the background by checking the characteristics of ridges of clusters on the joint histogram. 3.Determine the combinations of (I 1, I 2 ) covered by the clusters as insignificant changes, and determine the rest of the combinations as significant changes. 4.Build a table Sig(I 1, I 2 ) which classifies combinations of (I 1, I 2 ) into significant / insignificant changes. 5.Extract pixels with significant changes from the images by comparing (I 1 (x), I 2 (x)) with the table. 6.Remove small size regions from the candidates for significant changes. 7.Determine the change area of the image by checking gradient correlation between the images for the candidate regions.

9. Server 1.Download images from the image server to the image change detection server. 2.Execute the image change detection program, and calculate the value of change between images. 3.Transform the change information into XML format and output it to a file.

10. XML Data Format … <event start="2007/04/01 07:50:00 +0900" image="http://fsds.dc.affrc.go.jp/data4/ Ichikawa10_cam/200704/20070401/ 200704010750.jpg“ change="12932“ area="(69,207)-(180,302)" /> … URL of an image Value of change Coordinates of change area Time

11. Web Application (1) Field Server Date Display Type

12. Web Application (2) Thumbnail Previous ImageDetected Image Change Value & Time

13. Web Application (3) A Square means a Detected Change Area Previous ImageDetected Image

14. Web Application (4) Timeline Short span band Long span band Timeline is a DHTML-based Ajax widget for visualizing time-based events. Bands can be panned by dragging with the mouse. A band synchronizes with panning of another band. Previous ImageDetected Image

15. Results (1) Appearance of cars in the fruit farm and the vegetable field.

16. Results (2) Appearance of farmers in the vegetable field.

17. Incorrect Results The strong density change by daylight The movement of clouds

18. Conclusions / Future Works Expected role of this system To appeal the safety of agricultural products by extracting images of agricultural work and show them to a consumer To monitor the theft of agricultural products and illegal disposal of garbage. To raise the accuracy of the image change detection Setting up the threshold for an image change and optimizing the program parameters based on the installation location of a Field Server.