Tree Leaf Classification for a Mobile Field GuideDavid Knight, James Painter, Matthew Potter

Department of Electrical Engineering, Stanford University

Motivation Classification Techniques

Related Work

Experimental Results

Species identification for hikers, park rangers, backyard enthusiasts Automated sorting for large-scale

research applications Targeted crop spraying (e.g. weeds)

Aspect ratio Rectangularity Convex hull area ratio Convex hull

perimeter ratio Sphericity Circularity

Eccentricity Form factor Regional moment of

inertia (4-D) Angle code

histogram (5-D)

Class1 Class2 Class3 Class4 Class5 Class6 MarginClass1 0.31 0.89 2.44 2.18 2.11 1.71 187%Class2 1.43 0.17 2.23 2.02 1.91 1.49 741%Class3 2.53 2.12 0.53 1.93 1.82 0.88 66.0%Class4 2.49 2.18 2.05 0.39 0.85 1.61 118%Class5 2.41 2.07 2.10 1.01 0.32 1.55 216%Class6 2.36 1.91 1.15 1.90 1.84 0.24 379%

Leaf class similarity scores; margin is difference between best and second best matches (MATLAB-generated)

Challenges In-class variation Shape deformities

1 2 3 4 5 6

Match Rate

Class1 100%Class2 100%Class3 80%Class4 100%Class5 93%Class6 100%

Leaf class match counts, 90 leaves

Platform Motorola DROID Google Android 2.x

Features Real-time leaf contour overlay in viewfinder Touch-based image rotation Immediate analysis of images from built-in camera On-board processing for use in absence of network coverage

LeafR: Mobile App Implementation

Centroid-Contour Distance, Angle Code Histogram: Z. Wang, Z. Chi, D. Feng, “Shape based leaf image retrieval,” IEE Proc. Visual Image Signal Process., Vol. 150, No. 1, Feb. 2003.

Dimensionless Descriptors: J.-X. Du, X.-F. Wang, G.-J. Zhang, “Leaf shape based plant species recognition,” Applied Mathematics and Computation, Vol. 185, No. 2, 15 Feb. 2007.;

S. Yonekawa, N. Sakai, O. Kitani, “Identification of idealized leaf types using simple dimensionless shape factors by image analysis,” Transactions of the ASAE, Vol. 39, No. 4, April 1996.

Agriculture: J. Hemming, T. Rath, “Computer Vision based Weed Identification under Field Conditions using Controlled Lighting,” Journal of Agricultural Engineering Research, Vol. 78, No. 3, pp. 233, 2001.

Leaf Classifiers

)4(

)3(

)2(

)1(

XX

XX

XX

XX

IIII

Leaf Image

Compute Classifiers

Measure Classifier Similarity

Classifier-Space Euclidean Dist.

(17 dim.)

Trained Leaf Species Classes

Extract Contour Leaf class match scores, Class

1-6 vs. Class 1-6 vs. Score

Hemming, et al. 2001 Du, et al. 2007 Yonekawa, et al. 1996