Journal of Visual Communication and Image Representation
Automatic identification of fruit flies (Diptera: Tephritidae)
Introduction
Fruit flies are pests of major economic importance in agriculture among which we highlight some species of genus Anastrepha. Anastrepha species occur exclusively in the American tropics and subtropics, with more than 250 valid and numerous undescribed species [1], but fewer than 10 species are of agricultural importance [2]. It is the most diverse genus of fruit flies, and most of its species have been divided into several species groups. However, the taxonomy of some groups is not yet properly solved [3]. In the fraterculus group, are included major pest species such as Anastrepha fraterculus (South American fruit fly) and Anastrepha obliqua (West Indian fruit fly). Both species infest several host species throughout their distribution. For example, larvae of A. fraterculus and A. obliqua develop on 81 and 36 commercial and wild host species, respectively, in Brazil [4].
The extent of fruit fly damage to commercially produced fruit is significant. In addition, quarantine restrictions imposed by fruit importing countries are another serious economic impact caused in case one fruit flies to be found. In this scenario, species identification is crucial for the implementation of management, control programs, and quarantine restrictions.
Anastrepha species identification is mainly based on subtle differences in the shape of the aculeus (the female egg-laying “needle”), but thoracic markings, wing pattern, and microtrichia are also important taxonomically. In the fraterculus group, besides external morphology, molecular [5], genetic [6], and morphometric [7] studies have also been carried out to clarify the identity of cryptic species.
Given the demand, novel tools for a quick and precise identification of fruit flies, amenable to automation, need to be developed and tested. This demand for computational solutions is due to a constant quest for reducing the time and costs in performing identification tasks. Among existing solutions, there are image analysis and machine learning techniques, which have been widely used in several application areas (e.g., security, medical image analysis, biology, and agriculture). In applications to agriculture, the use of image analysis and machine learning techniques is not rare. Arriba et al. [8], for example, have proposed an automatic leaf image classification system for sunflower crops. In [9], [10], image processing techniques have been applied to classify or identify wheat, spelt, and hybrid seeds. In [11], in turn, texture analysis has been performed to differentiate bark from wood chips.
For automatic identification of species, some systems have been developed, such as: (1) Digital Automated Identification SYstem (DAISY), which performs fish, pollen, plant, and butterfly classification [12]; (2) SPecies IDentified Automatically (SPIDA-web), which is a tool for identifying Australian spiders, that allows to distinguish 121 species [13]; (3) Automatic Bee Identification System (ABIS), which identifies bees for species of genus Bombus, Colletes, and Andrena [14]. In this work, we used, for the first time in fruit flies, image analysis techniques to automatically identify three species of the fraterculus group: Anastrepha fraterculus (Wied.), Anastrepha obliqua (Macquart) and Anastrepha sororcula Zucchi through wings and aculei.
The objective of this work is to perform a robust study of description and learning techniques that can serve as support for the development of the first automatic identification system of these species using wing and aculeus images. We explore complementary image features through the use of a framework for classifier selection and fusion that point out which ones are more effective to capture the properties and nuances of the fruit flies allowing us to devise and deploy an automatic classification system. Furthermore, this work can serve as a guide for implementing new modules into existing systems in the literature.
We evaluated the use of several image description approaches that encode the color, texture, and shape properties of images of fruit fly wings and aculei into feature vectors. Those features are then used to train classifiers that are later combined by a meta-learning approach that identifies fruit fly species. In addition, we also assess the feasibility of classifying fruit flies based on their wings in order to make this task more objective than it is nowadays. In this context, the main contributions of this paper are:
- 1.
Evaluation of several image descriptors for classifying fruit flies.
- 2.
Exploration, for the first time, of fruit fly wings as a possible morphological feature for automatic classification.
- 3.
Design and development of an automatic classification and fusion system able to explore complementary features present in wings and aculei of fruit flies.
The remainder of this paper is organized as follows. Section 2 presents related concepts necessary for the understanding this paper. Section 3 describes the proposed classifier multimodal fusion framework. Section 4 shows the experimental procedure we adopted to conduct our experiments, while Section 5 discusses the experiments and results. Finally, Section 6 concludes the paper and points out future research directions.
Section snippets
Related work and background
This section shows important concepts used in this work.
A framework for classifier selection and fusion
Given a classification problem, we have a set of image descriptors and a set of simple learning methods that will be used to learn patterns from available images in the training set. The important question then is how to automatically find the best classifiers and image properties and, more importantly, how to combine them to achieve the best possible classification results.
In this work, we denote a classifier as a tuple formed by a learning method and an image descriptor. Once we train all
Experimental protocol
This section shows the setup for each step introduced in Section 2.1 and presents the experimental methodology adopted to validate this work.
Experiments and results
In this section, we present results obtained with each simple classifier and all three fusion approaches that we propose herein (see Section 3). Section 5.1 discusses the impact of the segmentation procedure employed, while Section 5.2 shows a detailed comparison of the best simple classifiers for each used image descriptor. Section 5.3 compares the classification performance of three fusion approaches and shows how the fusion outperforms any single method without it. Finally, Section 5.4 shows
Conclusions and future work
In this study, we performed several experiments with different image description and learning methods to develop a sounding understanding and basis for the design and deployment of a fruit fly recognition system considering the fraterculus complex. In these experiments, two datasets were used (WING and ACULEU), three approaches (FSVM-WINGS, FSVM-ACULEI, and FSVM-MULTIMODAL) have been proposed and different behaviors noticed. The first two fusion methods explore different classification and
Acknowledgments
We thank the financial support of the FAPESP (Grants 2010/14910-0, 2010/05647-4, 2009/54806-0, and 98/05085-2), CNPq (Grants 303726/2009-1, 550890/2007-6, 309618/2010-0, and 304352/2012-8), CAPES (Grant 1260–12-0), AMD and Microsoft. We also thank Miguel Francisco Souza Filho for the samples of Anastrepha, and Heraldo Negri de Oliveira for the fruit fly photo.
References (56)
- et al.
Leaf classification in sunflower crops by computer vision and neural networks
Comput. Electron. Agric.
(2011) - et al.
Identification of nine Iranian wheat seed varieties by textural analysis with image processing
Comput. Electron. Agric.
(2012) - et al.
Identification of hybrids of spelt and wheat and their parental forms using shape and color descriptors
Comput. Electron. Agric.
(2012) - et al.
Discrimination of bark from wood chips through texture analysis by image processing
Comput. Electron. Agric.
(2011) - et al.
Innovations in fingerprint capture devices
Pattern Recogn. (PR)
(2003) - et al.
A genetic programming framework for content-based image retrieval
Pattern Recogn. (PR)
(2009) - et al.
Automatic fruit and vegetable classification from images
Elsevier Comput. Electron. Agric. (COMPAG)
(2010) - et al.
A framework for selection and fusion of pattern classifiers in multimedia recognition
Pattern Recognit. Lett.
(2014) - et al.
Texture recognition and image retrieval using gradient indexing
J. Vis. Commun. Image Representat.
(2000) - et al.
Image retrieval based on shape similarity by edge orientation autocorrelogram
Pattern Recognit.
(2003)
Comparative study of global color and texture descriptors for web image retrieval
J. Vis. Commun. Image Representat.
Bionomics and management of Anastrepha
Annu. Rev. Entomol.
Phylogeny of the genera Anastrepha and Toxotrypana (Trypetinae: Toxotrypanini) based on morphology
Phylogenetic relationships among species of the fraterculus group (Anastrepha: Diptera: Tephritidae) inferred from DNA sequences of mitochondrial cytochrome oxidase I
Neotrop. Entomol.
A genetic-morphological characterization of two cryptic species of the Anastrepha fraterculus complex (Diptera: Tephritidae)
Ann. Entomol. Soc. Am.
Cryptic species of the Anastrepha fraterculus complex (Diptera, Tephritidae): a multivariate approach for the recognition of South American morphotypes
Ann. Entomol. Soc. Am.
A qualitative study investigating automated identification of living macrolepidoptera using the Digital Automated Identification SYstem (DAISY)
Syst. Biodivers.
Introducing SPIDA-web: wavelets, neural networks and Internet accessibility in an image-based automated identification system
Syst. Assoc.
Morphometric analysis of Mexican and South American populations of the Anastrepha fraterculus complex (Diptera: Tephritidae) and recognition of a distinct Mexican morphotype
Bull. Entomol. Res.
A survey of cost-sensitive decision tree induction algorithms
ACM Comput. Surv.
Introduction to Data Mining
Induction of decision trees
Mach. Learn.
C4.5: Programs for Machine Learning
The Elements of Statistical Learning
Pattern Classification
Cited by (41)
Fruit fly management research: A systematic review of monitoring and control tactics in the world
2018, Crop ProtectionCitation Excerpt :Brazilian researchers implemented a classifier multimodal fusion approach, using two types of images (wings and aculei), generating promising results for the identification of Anastrepha species. The results showed more than 98% classification accuracy, which is remarkable, despite the technical problems (Faria et al., 2014). The risk of not detecting early or not responding immediately to the detections of exotic fruit flies can be illustrated by cases where eradication failed, such as B. carambolae in Suriname.
Deep Learning Based Classification for Hoverflies (Diptera: Syrphidae)
2023, Journal of the Entomological Research SocietyA Comparative Study of Deep Convolutional Neural Network Architectures to Identify Full Bee Body in Images
2023, International Conference on Information and Digital Technologies 2023, IDT 2023An IoT Based Helopeltis Sp Pest Control System
2023, Communications in Computer and Information ScienceDeep Learning Embedded into Smart Traps for Fruit Insect Pests Detection
2022, ACM Transactions on Intelligent Systems and Technology