Identification of Costa Rican Plant Species using Computer Vision

Detalles

Categoría: Project Identification of Costa Rican Plant Species using Computer Vision

Develop Computer Vision algorithms to automatically identify plant species of Costa Rica based purely on images of leaves. including leaf segmentation, shadow removal, noise removal, feature extraction and similarity search.

Click here to see video

It is estimated that the amount of plant species in our planet is around 400,000. Costa Rica alone is expected to have about 12,000 species of plants, of which approximately 11,000 species have already been identified. Overall, plant identification is fundamental for biological richness studies of a region, plant and endangered species population monitoring, climate change impact on forests, invasive species distribution modeling, and bioliteracy activities in rural and urban areas, among others.

Costa Rica has an extraordinary biodiversity (about 4% of the world's biodiversity) and has been recognized both because of the innovative use of ICT’s  to conserve biodiversity and for its protected areas system, which is a model worldwide. Nevertheless, plant species identification is a process that normally requires expert knowledge and use of dichotomous keys, interactive keys, or just the experience of an expert. This makes the process tedious, inefficient and error prone. 

We aim to support the efficient, automatic identification of plant species of Costa Rica by developing Computer Vision algorithms for plant leaf images analysis.

• Create a Costa Rican Leaf Image Database that includes all species that have been inventoried in the country.
• Extend the identification algorithms based on curvature and texture of the leaf, to include circularity detection and morphological measures used by taxonomists.
• Eliminate shadows from leaf images.
• Develop algorithms to extract/segment leaf pixels from herbaria samples.
• Develop algorithms to separate the leaf texture in its atomic elements such as venation, margin, porosity, to understand the classification power of each one.
• Determine the computational limits of computer vision based plant species identification. What is the accuracy per taxonomic group? (Perhaps per families) or other thematic groups such as endangered species.
• Create a mobile app that uses the developed algorithms plus geo referencing to identify species.

Pictures

 

Local Binary Patterns with different radius and sampling of a leaf texture.

 

Contours detection of a leaf image.

 

Curvature over scale of the leaf contour, based on the intersection of a disk with the leaf area at a given contour pixel.

 

HSV color domain splitting to avoid noise prior to leaf segmentation.

 

 Several leaf shapes segmented into binary images for posterior feature extraction.

 

Dr. Erick Mata-Montero, Advisor

Email: Esta dirección de correo electrónico está siendo protegida contra los robots de spam. Necesita tener JavaScript habilitado para poder verlo.

Phone: 2550-2115

Jose Carranza-Rojas

Email: Esta dirección de correo electrónico está siendo protegida contra los robots de spam. Necesita tener JavaScript habilitado para poder verlo.