1 points based on feature values or

1 IntroductionDeep learning is a speci c sub eld of machine learning that attempts to learn hierarchical abstractionsin input data with multiple layers of non-linear information processing representation 2, 7. In thiscontext, it is not a deep understanding archived by the approach but stands for this idea of successivelayers of representations. Although machine learning techniques have been been widely applied in datamining for many years, their ability to process natural and complex data (sound, image and video les) intheir raw form are limited 12. Other approaches to machine learning, sometimes called shallow learning,tend to focus on learning only one or two layers of representation spaces of the data usually throughsimple transformations such as high-dimensional non-linear projections (support vector machines) ordecision trees. Support vector machines are a classi cation technique that nd good decision boundariesbetween two sets of points belonging to two di erent categories and decision trees are owchart-likestructures that let us classify input data points based on feature values or predict output values giveninputs 11.

Deep learning has become an emerging approach in recent years with the advancement of hardwaretechnology (high-speed Central Processing Units (CPUs) and Graphic Processing Units (GPUs)), bigdata and algorithmic advances (deep neural network (DNN), convolutional neural network (CNN) andrecurrent neural networks). In deep learning, these multiple layered representations are learned viamodels called neural networks, structured in literal layers stacked on top of each other. A CNN generallyconsists of three main neural layers (convolutional layers, pooling layers, and fully connected layers) 8.Figure 1 illustrates an architecture of a typical convolutional neural network (CNN) which is structuredas a series of stages. An essential advantage of a deep CNN is the automatic learning of task-speci crepresentations of the input data which replace traditional feature-based representations using handcrafted features 1.

Sometimes it is hard to do all the work on your own
Let us help you get a good grade on your paper. Get expert help in mere 10 minutes with:
  • Thesis Statement
  • Structure and Outline
  • Voice and Grammar
  • Conclusion
Get essay help
No paying upfront

Image-based methods are considered a promising approach for species identi cation 5, 10. Inecology, several workers have used CNN in plant identi cation using vein morphological patterns 6,via optimization of transfer learning parameters 16 and taxon identi cation 1. Other more advancedmethods used in plant classi cation are discriminative deep feature learning 24, 20 and very deepconvolutional networks by incorporating species and organ features to solve the multi-organ plant classi cation problem 18, 15, 19.During the time of extraordinary levels of global human activity, many plant species have escapedoutside their homes to new lands. Their impacts on their foreign environments are diverse but mostlyharmful such as competition with native species on soil conditions, and they become invasive species 3.In order to successfully control invasive plants, it is imperative that non-experts be able to identify andreport them so that they may be removed swiftly.

Ecologists are hosting a Kaggle competition to seeif computers can accurately identify invasive species in photos. It has been found that this problem isvery tractable with deep CNNs.2 ObjectivesThe main objective of this research is to apply some deep learning algorithms to analyse some imagedatasets in ecology (invasive plant species and plant seedlings).1Figure 1: Representation of deep learning for face image analysis.3 Literature ReviewIn this section, we briey introduce how deep learning work and then review studies relating to itsapplication in ecology.3.1 How deep learning work?As mentioned in the Introduction section, machine learning is about mapping inputs (such as images) totargets (such as the label cat”).

This is done by observing many examples of input and targets. On theother hand, deep neural networks do this input-to-target mapping via a deep sequence of simple datatransformations (layers). These data transformations are learned by exposure to examples (Figure 2).The speci cation of what a layer does to its input data is stored in the layer’s weights, which inessence are a bunch of numbers. The transformation implemented by a layer is parameterised by itsweights which are also sometimes called the parameters of a layer.) In this context, learning means nding a set of values for the weights of all layers in a network, such that the network will correctly mapexample inputs to their associated targets. The problem is a deep neural network can contain a lot ofparameters (sometimes millions).

Finding the correct value for all of them may be a hide-and-seek game,especially given that modifying the value of one parameter will a ect the behavior of all the others.To control the output of a neural network, we need to be able to measure how far this output isfrom what we expected. This is the job of the loss function of the network. The loss function takes thepredictions of the network and the true target and computes a distance score, capturing how well thenetwork has done on this speci c example. The fundamental trick in deep learning is to use this score asa feedback signal to adjust the value of the weights a little, in a direction that will lower the loss scorefor the current example (see Figure 2). This adjustment is the job of the optimizer, which implementswhat is called the Backpropagation algorithm: the central algorithm in deep learning.3.

2 Deep learning techniques used in ecologyTaxon identi cation is an important step in many plant ecological studies. Plant identi cation systemsdeveloped by computer vision researchers have helped ecologists to recognise and identify unknown plantspecies more rapidly 13. Recent work by 22 is the rst systematic literature review with the aim ofa thorough analysis and comparison of primary studies on computer vision approaches for plant speciesidenti cation.2Figure 2: Representation of a deep learning algorithm.Mehdipour Ghazi et al. 16 used deep CNNs to identify the plant species captured in a photographand evaluate di erent factors a ecting the performance of these networks. Three powerful and populardeep learning architectures, namely GoogLeNet, AlexNet, and VGGNet, are used for this purpose.

Transfer learning is used to ne-tune the pre-trained models, using the plant task datasets of LifeCLEF2015. The most important factor a ecting ne-tuning performance is the number of iterations whiledata augmentation comes second. On the other hand, while increasing the batch size improves accuracy,increasing the number of iterations is a better use of computation time, considering the time complexityversus performance improvements. The observations collected in this work will shed some light to othersimilar visual recognition problems.Another successful study by Barre et al.

1 is LeafNet, a computer vision system for automatic plantspecies identi cation in which eciency and reproducibility might greatly bene t from partly automatingthis task. Image-based identi cation systems exist, but mostly rely on hand-crafted algorithms to extractsets of features chosen a priori to identify species of selected taxa. In consequence, such systems arerestricted to these taxa and additionally require involving experts that provide taxonomical knowledgefor developing such customized systems. The aim of this study was to develop a deep learning system tolearn discriminative features from leaf images along with a classi er for species identi cation of plants.

By comparing our results with customized systems like LeafSnapwe can show that learning the featuresby a CNN can provide better feature representation for leaf images compared to hand-crafted features.Lee et al. 14 presented a CNN approach to taxon identi cation based on leaf images and reported anaverage accuracy of 99.7% on a dataset covering 44 species.

Zhang et al. 23 used CNNs to classify theFlavia dataset with data augmentation and obtained an accuracy of 94.69%. Reyes et al. 17 pre-traineda convolutional neural network using 1.

8 million images and used a ne-tuning strategy to transfer learnedrecognition capabilities from general domains to the speci c challenge of Plant Identi cation task with3an average accuracy of 48.6%.Lee et al. 13 investigated the use of deep learning to harvest discriminatory features from leafimages by learning, and apply them as classi ers for plant identi cation. They found that classi cationperformance is a ected when constraining the varieties of leaf data to be seen by CNN during training.

Combining one global network trained upon the whole leaf data and another local network trained uponits corresponding leaf patches at the early stage is more bene cial as features are learned end-to-end,starting from before and after fusion. Therefore, experimentally that hybrid local-global features learnedusing deep learning can improve recognition performance compared to previous techniques.Jordan and Ian 21 developed an open-source deep learning software for image-based plant phenotyping called Deep Plant Phenomics. Their application provides pre-trained neural networks for severalcommon plant phenotyping tasks such as leaf counting task and set baseline results for the mutant classi cation and age regression tasks. This is an easy platform that can be used by plant scientists to trainmodels for their own phenotyping applications.

Zuo and Wang 24 indicated that by combining discriminative information with the deep ISA featurelearning framework it can signi cantly improve the performance of recognizing similar tree species ata distance. Sun et al. 20 developed a 26-layer deep learning model consisting of 8 residual buildingblocks is designed for large-scale plant classi cation in natural environment and their model achieves arecognition rate of 91.

78% on the BJFU100 dataset, demonstrating that deep learning is a promisingtechnology for smart forestry.Most other plant classi cation problems, such as in 14, 17 and 4 involve full-frame photos ofthe plant leaf with the background removed, rather than in a natural environment. One paper thatinvestigated classifying plants in their natural environment used residual networks of various depths 20but the plants were usually always in full-frame. Jobson and Hernandez 9 used a dataset containingmany images where hydrangeas only appear at the margins or are occluded. Few papers feature a needfor classi cation without regional detection, so the literature on this subject is limited.

The project will focus on exploring CNN models for the plant classi cation task including its implication in the open-set recognition task which requires learning new models that are available and gooddata les.References1 Pierre Barre, Ben C. Stover, Kai F. Muller, and Volker Steinhage. LeafNet: A computer visionsystem for automatic plant species identi cation. Ecological Informatics, 2017.

2 B. Chandra and Rajesh K. Sharma. Fast learning in deep neural networks. Neurocomputing, 171:1205{ 1215, 2016.3 Je rey D.

Corbin and Carla M. D’Antonio. Gone but not forgotten? invasive plants’ legacies oncommunity and ecosystem properties.

Invasive Plant Science and Management, 5(1):117{124, 2012.4_Ilke C ugu, Eren Sener, C agr Erciyes, Burak Balc, Emre Akn, ItrOnal, and Ahmet Oguz Akyuz.Treelogy: A novel tree classi er utilizing deep and hand-crafted representations. arXiv preprintarXiv:1701.08291, 2017.5 K.

J. Gaston and M. A. O’Neill. Automated species identi cation: why not? Philosophical Transactions of the Royal Society B: Biological Sciences, 2004.

6 Guillermo L. Grinblat, Lucas C. Uzal, Monica G. Larese, and Pablo M. Granitto.

Deep learning forplant identi cation using vein morphological patterns. Computers and Electronics in Agriculture,2016.7 Yanming Guo, Yu Liu, Ard Oerlemans, Songyang Lao, Song Wu, and Michael S. Lew.

Deep learningfor visual understanding: A review. Neurocomputing, 187:27 { 48, 2016. Recent Developments onDeep Big Vision.8 Yanming Guo, Yu Liu, Ard Oerlemans, Songyang Lao, Song Wu, and Michael S. Lew.

Deep learningfor visual understanding: A review. Neurocomputing, 2016.49 Elliott Jobson and Andres Hernandez. Visual Classi er for Invasive Plant Species.10 Alexis Joly, Henning Muller, Herve Goeau, Herve Glotin, Concetto Spampinato, Andreas Rauber,Pierre Bonnet CIRAD, Willem-Pier Vellinga, and Bob Fisher.

LifeCLEF: Multimedia Life SpeciesIdenti cation. 2014.11 Sotiris B Kotsiantis, Ioannis D Zaharakis, and Panayiotis E Pintelas.

Machine learning: a reviewof classi cation and combining techniques. Arti cial Intelligence Review, 26(3):159{190, 2006.12 Yann LeCun, Yoshua Bengio, and Geo rey Hinton. Deep learning. nature, 521(7553):436{444, 2015.13 Sue Han Lee, Chee Seng Chan, Simon Joseph Mayo, and Paolo Remagnino. How deep learningextracts and learns leaf features for plant classi cation.

Pattern Recognition, 2017.14 Sue Han Lee, Chee Seng Chan, Paul Wilkin, and Paolo Remagnino. Deep-plant: Plant identi cationwith convolutional neural networks. In Proceedings – International Conference on Image Processing,ICIP, 2015.

15 Sue Han Lee, Yang Loong Chang, Chee Seng Chan, and Paolo Remagnino. Plant identi cationsystem based on a convolutional neural network for the lifeclef 2016 plant classi cation task. InCEUR Workshop Proceedings, 2016.16 Mostafa Mehdipour Ghazi, Berrin Yanikoglu, and Erchan Aptoula.

Plant identi cation using deepneural networks via optimization of transfer learning parameters. Neurocomputing, 2017.17 Angie K.

Reyes, Juan C. Caicedo, and Jorge E. Camargo. Fine-tuning deep convolutional networksfor plant recognition. In CEUR Workshop Proceedings, 2015.18 Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale imagerecognition.

2015.19 MilanSulc, Dmytro Mishkin, and Jir Matas. Very deep residual networks with MaxOut for plantidenti cation in the wild. In CEUR Workshop Proceedings, 2016.20 Yu Sun, Yuan Liu, Guan Wang, and Haiyan Zhang.

Deep Learning for Plant Identi cation inNatural Environment. Computational Intelligence and Neuroscience, 2017.21 Jordan R. Ubbens and Ian Stavness. Deep Plant Phenomics: A Deep Learning Platform for ComplexPlant Phenotyping Tasks. Frontiers in Plant Science, 2017.22 Jana Waldchen and Patrick Mader. Plant Species Identi cation Using Computer Vision Techniques:A Systematic Literature Review.

Archives of Computational Methods in Engineering, 0, 123.23 Chaoyun Zhang, Pan Zhou, Chenghua Li, and Lijun Liu. A Convolutional Neural Network forLeaves Recognition Using Data Augmentation.24 Zhen Zuo and Gang Wang. Recognizing trees at a distance with discriminative deep feature learning.In ICICS 2013 – Conference Guide of the 9th International Conference on Information, Communications and Signal Processing, 2013.5


I'm Gerard!

Would you like to get a custom essay? How about receiving a customized one?

Check it out