A Novel Leaf-fragment Dataset and ResNet for Small-scale Image Analysis
Download as PPTX, PDF0 likes24 views
This document presents research on using deep learning for plant identification through leaf vein patterns. The researchers collected over 80,000 leaf images of 4 plant species and 2 cotyledon types. They extracted the green color channel and center cropped the images. Various image sizes were then generated for analysis. A ResNet model was used for cotyledon and species classification and compared to ResNet-152. The model achieved good performance on the small dataset. Future work could include adding more image data and exploring other color channels.
![Related works
Name of the paper Used models Dataset Performance
Deep Learning For plant
identification using vein
morphological patterns [3]
Deep CNN 208 images of white bean, red bean
and soybean plants.
Setup S1 - 93.0+-0.3 (CNN)
Setup S2 - 96.9+-0.2 (CNN)
Deep Learning for plant species
classification using leaf vein
morphometric [1]
CNN, SVM, ANN, K-NN, NB 43 different species plants with 1,290
leaf images.
ANN classifier gained the best
accuracy with 95.54%
An implementation of leaf
recognition system using leaf vein
and shape [16]
Classification using features like
leaf-vein and shape.
1,907 images plant leaf images of 32
different species
97.19% .
Partial differential equations and
fractal analysis to plant leaf
identification [10]
Bouligand-Minkowski , PDE,
Hybrid method
40 species of plants with 20 leaves for
per species.
87.00 (±1.11)
Automatic classification of legumes
using leaf vein image features [7]
SVM (with linear and Gaussian
kernels), penalized discriminant
analysis and random forests
422 images correspond to soybean
leaves, 272 images to red bean leaves
and 172 to white bean leaves.
PDA 87.37% for scanned leaves &
89.17% for cleared leaves
Morphological features for leaf-
based plant recognition [6]
Two descriptors based on
mathematical morphology;
morphological covariance &
covariance histogram. 9 percentile
points w.r.t the CCH operator
were used
ImageClef’12 plant identification
dataset.
n ECCH+ACH+CPDH+CC provides
the best result with Accuracy
56.09% and σ with 3.12.](https://image.slidesharecdn.com/iciss005-210504061448/85/A-Novel-Leaf-fragment-Dataset-and-ResNet-for-Small-scale-Image-Analysis-3-320.jpg)
A Novel Leaf-fragment Dataset and ResNet for Small-scale Image Analysis
- 1. Abdul Hasib Uddin Presented By Authors and affiliations Abdul Hasib Uddin1,*, Sharder Shams Mahamud2, Abu Shamim Mohammad Arif3 Computer Science and Engineering Discipline, Khulna University, Khulna, Bangladesh. Email: 1abdulhasibuddin@gmail.com, 2info.shamsbd@gmail.com, 3shamimarif@yahoo.com
- 2. Introduction & Motivation There are approximately 391,000 vascular species all around the world. Therefore, it is inconceivable and impractical for a specialist, to be able to recognize and categorize all the species Plant identification is not exclusively the job of botanists and plant ecologists. It is useful and even sometimes required in numerous parts of society, from professionals (such as landscape architects, foresters, farmers, conservationists, and biologists) to the general public (like ecotourists, hikers, and nature lovers). Nonetheless, the identification of plants by conventional means is difficult, time consuming, and frustrating for novices.
- 3. Related works Name of the paper Used models Dataset Performance Deep Learning For plant identification using vein morphological patterns [3] Deep CNN 208 images of white bean, red bean and soybean plants. Setup S1 - 93.0+-0.3 (CNN) Setup S2 - 96.9+-0.2 (CNN) Deep Learning for plant species classification using leaf vein morphometric [1] CNN, SVM, ANN, K-NN, NB 43 different species plants with 1,290 leaf images. ANN classifier gained the best accuracy with 95.54% An implementation of leaf recognition system using leaf vein and shape [16] Classification using features like leaf-vein and shape. 1,907 images plant leaf images of 32 different species 97.19% . Partial differential equations and fractal analysis to plant leaf identification [10] Bouligand-Minkowski , PDE, Hybrid method 40 species of plants with 20 leaves for per species. 87.00 (±1.11) Automatic classification of legumes using leaf vein image features [7] SVM (with linear and Gaussian kernels), penalized discriminant analysis and random forests 422 images correspond to soybean leaves, 272 images to red bean leaves and 172 to white bean leaves. PDA 87.37% for scanned leaves & 89.17% for cleared leaves Morphological features for leaf- based plant recognition [6] Two descriptors based on mathematical morphology; morphological covariance & covariance histogram. 9 percentile points w.r.t the CCH operator were used ImageClef’12 plant identification dataset. n ECCH+ACH+CPDH+CC provides the best result with Accuracy 56.09% and σ with 3.12.
- 4. Our Contributions Our Contibutions Created center-focused green- channel leaf image dataset Used “only” vein patterns for identifying plant species Made small scale dataset instances (64x64p, 32x32p, 16x16p, 8x8p, and 4x4p) Applyed proposed ResNet model, investigated performances, and compared with ResNet-152 V2
- 5. Proposed method Data collection: Hardware augmentation. Cotyledon: 2 Species: 4 Image: 81,527 Image Pre-processing: i) Green-Channel extraction ii) Pulling image center Post processing: i) Down sampling (images: 43,656) ii) Array conversion Model Implementation: i) ResNet for cotyledon- type identification ii) ResNet for species classification Compare the performance Fig: Block schematic of the work flow.
- 10. Results
- 11. Conclusion & Future work In this research, we introduced a novel processed center-focused green- channel image dataset for cotyledon-type identification and plant species classification. Multiple instances of the dataset were created based on image dimension and shown the effects of image dimension on Residual Neural Network. Instead of down-sampling the dataset from each group, they can be either up- sampled using Random Sampling with Repeat or more images in each category can be added. Red and blue channels can be extracted from the images to observe the impacts.
- 12. References 1. Tan, Jing Wei, Siow-Wee Chang, Sameem Binti Abdul Kareem, Hwa Jen Yap, and Kien-Thai Yong. "Deep learning for plant species classification using leaf vein morphometric." IEEE/ACM transactions on computational biology and bioinformatics (2018). 2. Lee, Sue Han, Chee Seng Chan, Simon Joseph Mayo, and Paolo Remagnino. "How deep learning extracts and learns leaf features for plant classification." Pattern Recognition 71 (2017): 1-13. 3. Grinblat, Guillermo L., Lucas C. Uzal, Mónica G. Larese, and Pablo M. Granitto. "Deep learning for plant identification using vein morphological patterns." Computers and Electronics in Agriculture 127 (2016): 418-424. 4. Walls, Ramona L. "Angiosperm leaf vein patterns are linked to leaf functions in a global‐scale data set." American journal of botany 98, no. 2 (2011): 244-253. 5. Bruno, Odemir Martinez, Rodrigo de Oliveira Plotze, Mauricio Falvo, and Mário de Castro. "Fractal dimension applied to plant identification." Information Sciences 178, no. 12 (2008): 2722-2733. 6. Hickey, Leo J. "Classification of the architecture of dicotyledonous leaves." American journal of botany 60, no. 1 (1973): 17-33. 7. Larese, Mónica G., Rafael Namías, Roque M. Craviotto, Miriam R. Arango, Carina Gallo, and Pablo M. Granitto. "Automatic classification of legumes using leaf vein image features." Pattern Recognition 47, no. 1 (2014): 158-168. 8. Kadir, Abdul, Lukito Edi Nugroho, Adhi Susanto, and Paulus Insap Santosa. "Leaf classification using shape, color, and texture features." arXiv preprint arXiv:1401.4447 (2013). 9. Aptoula, Erchan, and B. Yanikoglu. "Morphological features for leaf based plant recognition." In 2013 IEEE International Conference on Image Processing, pp. 1496-1499. IEEE, 2013. 10. Machado, Bruno Brandoli, Dalcimar Casanova, Wesley Nunes Gonçalves, and Odemir Martinez Bruno. "Partial differential equations and fractal analysis to plant leaf identification." In Journal of Physics: Conference Series, vol. 410, no. 1, p. 012066. 2013. 11. Price, Charles A., Scott Wing, and Joshua S. Weitz. "Scaling and structure of dicotyledonous leaf venation networks." Ecology letters 15, no. 2 (2012): 87-95. 12. Shakya, Subarna. "Analysis of Artificial Intelligence based Image Classification Techniques." Journal of Innovative Image Processing (JIIP) 2, no. 01 (2020): 44-54. 13. Jacob, I. Jeena. "Performance Evaluation of Caps-Net Based Multitask Learning Architecture for Text Classification." Journal of Artificial Intelligence 2, no. 01 (2020): 1-10. 14. Chandy, Abraham. "Pest Infestation Identification in Coconut Trees Using Deep Learning." Journal of Artificial Intelligence and Capsule Networks 1, no. 1 (2019): 10-18. 15. Vijayakumar, T., and Mr R. Vinothkanna. "Mellowness Detection of Dragon Fruit Using Deep Learning Strategy." Journal of Innovative Image Processing (JIIP) 2, no. 01 (2020): 35-43. 16. Lee, Kue-Bum, and Kwang-Seok Hong. "An implementation of leaf recognition system using leaf vein and shape." International Journal of Bio-Science and Bio-Technology 5, no. 2 (2013): 57-66. 17. Wu, Stephen Gang, Forrest Sheng Bao, Eric You Xu, Yu-Xuan Wang, Yi-Fan Chang, and Qiao-Liang Xiang. "A leaf recognition algorithm for plant classification using probabilistic neural network." In 2007 IEEE international symposium on signal processing and information technology, pp. 11-16. IEEE, 2007. 18. He, Kaiming, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. "Deep residual learning for image recognition." In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770-778. 2016. 19. He, Kaiming, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. "Identity mappings in deep residual networks." In European conference on computer vision, pp. 630-645. Springer, Cham, 2016.