Search

Computer vision (CV) has been applied for years to automate many human activities. It is one of the key technologies for the modernization of the agri-food industry towards the fourth industrial revolution (Industry 4.0). In the agricultural sector, CV systems are applied to automate or obtain information from many agricultural tasks such as planting, cultivation, farm management, disease control, weed control or robotic harvesting. It is also widely used in postharvest to automate and obtain objective information in processes such as quality control and evaluation, damage detection, classification of fruits or vegetables in commercial categories or composition analysis. One of the main advantages is the ability of this technology to obtain information in regions of the spectrum that are invisible to the human eye. An example is the case of hyperspectral imaging systems. These systems generate a large amount of data that needs to be processed efficiently, creating robust and repeatable statistical models that allow the technology to be implemented at an industrial level. To achieve this, it is necessary to couple CV systems with advanced artificial intelligence tools such as machine learning or deep learning. The objective of this work is to review the latest advances in CV systems applied to food and agricultural products and processes. RESUMO A visão computacional (CV) tem sido aplicada há anos para automatizar muitas atividades humanas. É uma das tecnologias-chave para a modernização da indústria agroalimentar em direção à quarta revolução industrial (Indústria 4.0). No setor agrícola, sistemas CV são aplicados para automatizar ou obter informações de muitas tarefas agrícolas, como plantio, cultivo, gerenciamento de fazenda, controle de doenças, controle de ervas daninhas ou colheita robótica. Também é amplamente utilizado em pós-colheita para automatizar e obter informações objetivas em processos como controle de qualidade e avaliação , detecção de danos, classificação de frutas ou vegetais em categorias comerciais ou análise de composição. Uma das principais vantagens é a capacidade desta tecnologia de obter informações em regiões do espectro invisíveis ao olho humano. Um exemplo é o caso de sistemas de imagens hiperespectrais. Esses sistemas geram uma grande quantidade de dados que precisam ser processados de forma eficiente, criando modelos estatísticos robustos e repetíveis que permitam a tecnologia a ser implementada a nível industrial. Para isso, é necessário acoplar os sistemas de CV a ferramentas avançadas de inteligência artificial, como aprendizado de máquina ou aprendizado profundo. O objetivo deste trabalho é revisar os últimos avanços em sistemas de CV aplicados a alimentos e produtos e processos agrícolas.

This chapter provides an overview of the basic concepts of color imaging and image processing techniques applied to sensing, monitoring and robotic operations in agriculture. To obtain good results with a vision system, it is very important to acquire high-quality images, particularly when captured with moving platform in a natural environment, by selecting a proper camera, acquisition settings, and lighting conditions. Image acquisition using CMOS and CCD sensors are explained along with proper adjustment of various imaging parameters such as aperture and shutter speed. Various color models that are relevant to image processing are described including RGB, HSV, HLS, CIELAB, and CIELUV as well as conversions between different color models. Following the introduction of color models, some basic image preprocessing techniques including image enhancement using histograms, morphological operations, and lowpass filtering are described. Also, various segmentation methods are discussed such as pixelwise or region–based techniques and classifiers. In addition, the chapter describes different object detection methods (with examples) that utilize various features such as colors, shapes, and textures. Hough transform and pattern matching are also commonly used techniques to detect various objects, and example applications based on these techniques are discussed. Finally, some of the crucial challenges for outdoor imaging such as varying illumination, occlusion, clustering, and movement of either the object or the camera when it is installed on a ground robotic system are discussed and a brief thought on future direction around these topics is presented.

[EN] Product inspection is essential to ensure good quality and to avoid fraud. New nectarine cultivars with similar external appearance but different physicochemical properties may be mixed in the market, causing confusion and rejection among consumers, and consequently affecting sales and prices. Hyperspectral reflectance imaging in the range of 450¿1040 nm was studied as a non-destructive method to differentiate two cultivars of nectarines with a very similar appearance but different taste. Partial least squares discriminant analysis (PLS-DA) was used to develop a prediction model to distinguish intact fruits of the cultivars using pixel-wise and mean spectrum approaches, and then the model was projected onto the complete surface of fruits allowing visual inspection. The results indicated that mean spectrum of the fruit was the most accurate method, a correct discrimination rate of 94% being achieved. Wavelength selection reduced the dimensionality of the hyperspectral images using the regression coefficients of the PLS-DA model. An accuracy of 96% was obtained by using 14 optimal wavelengths, whereas colour imaging and a trained inspection panel achieved a rate of correct classification of only 57% of the fruits., This work was partially funded by INIA and FEDER funds through project RTA2015-00078-00-00. Sandra Munera thanks INIA for the FPI-INIA grant num. 43 (CPR2014-0082), partially supported by European Union FSE funds. The authors wish to thank Fruits de Ponent (Lleida) for providing the fruit.

Anthracnose is one of the primary diseases that affect olive production before and after harvest, causing severe damage and economic losses. The objective of this work is to detect this disease in the early stages, using hyperspectral images and advanced modelling techniques of Deep Learning (DL) and convolutional neural networks (CNN). The olives were artificially inoculated with the fungus. Hyperspectral images (450–1050 nm) of each olive were acquired until visual symptoms of the disease were observed, in some cases up to 9 days. The olives were classified into two classes: control, inoculated with water, and fungi composed of olives inoculated with the fungus. The ResNet101 architecture was chosen and adapted to process 61-band hyperspectral images with only two classes. The result showed that the applied model is very effective in detecting infected olives since the sensitivity of the method was very high from the beginning (85% on day 3 and 100% onwards). From a commercial point of view, these results align with the need to detect the maximum number of infected fruits.

Conventional analytical techniques employ destructive methods, which are normally expensive, contaminating, time-consuming and only a few samples per batch can be monitored at a time. Hyperspectral imaging, instead, can be applied to the inspection of a large range of food, including fish, meat, fruit, and vegetables. Depending on the type of food and the properties to be analyzed, different wavelength dispersion devices, cameras, or illumination sources have to be used to capture images in the most appropriate spectral ranges. Later, specific statistical prediction or classification models have to be built to analyze the huge amount of data captured by such devices. This chapter explores the use of hyperspectral imaging for practical applications in food quality and safety inspection. Different technologies for acquiring the images as well as the most commonly used methods to extract useful information from the images are described by analyzing the most recent applications.

Loquat (Eriobotrya japonica L.) is an important fruit for the economy of some regions of Spain that is very susceptible to mechanical damage and physiological disorders. These problems depreciate its value and prevent it from being exported. Visible (VIS) and near infrared (NIR) hyperspectral imaging was used to discriminate between external and internal common defects of loquat cv. ‘Algerie’. Two classifiers, random forest (RF) and extreme gradient boost (XGBoost), and different spectral pre-processing techniques were evaluated in terms of their capacity to distinguish between sound and defective features according to three approaches. In the first approach the fruit pixels were classified into two classes, sound or defect, with a 97.5% rate of success; in the second the defective features were considered internal or external defects, achieving a 96.7% rate of success; and in the third approach each type of defect, i.e. purple spot, bruising, scars and flesh browning, were considered separately with a correct classification rate of 95.9%. The results indicated that the XGBoost classifier was the best method in all cases.

BACKGROUND: Grapevine cluster morphology influences the quality and commercial value of wine and table grapes. It is routinely evaluated by subjective and inaccurate methods that do not meet the requirements set by the food industry. Novel two-dimensional (2D) and three-dimensional (3D) machine vision technologies emerge as promising tools for its automatic and fast evaluation. RESULTS: The automatic evaluation of cluster length, width and elongation was successfully achieved by the analysis of 2D images, significant and strong correlations with the manual methods being found (r = 0.959, 0.861 and 0.852, respectively). The classification of clusters according to their shape can be achieved by evaluating their conicity in different sections of the cluster. The geometric reconstruction of the morphological volume of the cluster from 2D features worked better than the direct 3D laser scanning system, showing a high correlation (r = 0.956) with the manual approach (water displacement method). In addition, we constructed and validated a simple linear regression model for cluster compactness estimation. It showed a high predictive capacity for both the training and validation subsets of clusters (R2 = 84.5 and 71.1%, respectively). CONCLUSION: The methodologies proposed in this work provide continuous and accurate data for the fast and objective characterisation of cluster morphology. © 2016 Society of Chemical Industry. © 2016 Society of Chemical Industry

RobHortic is a remote-controlled field robot that has been developed for inspecting the presence of pests and diseases in horticultural crops using proximal sensing. The robot is equipped with colour, multispectral, and hyperspectral (400&ndash, 1000 nm) cameras, located looking at the ground (towards the plants). To prevent the negative influence of direct sunlight, the scene was illuminated by four halogen lamps and protected from natural light using a tarp. A GNSS (Global Navigation Satellite System) was used to geolocate the images of the field. All sensors were connected to an on-board industrial computer. The software developed specifically for this application captured the signal from an encoder, which was connected to the motor, to synchronise the acquisition of the images with the advance of the robot. Upon receiving the signal, the cameras are triggered, and the captured images are stored along with the GNSS data. The robot has been developed and tested over three campaigns in carrot fields for the detection of plants infected with &lsquo, Candidatus Liberibacter solanacearum&rsquo, The first two years were spent creating and tuning the robot and sensors, and data capture and geolocation were tested. In the third year, tests were carried out to detect asymptomatic infected plants. As a reference, plants were analysed by molecular analysis using a specific real-time Polymerase Chain Reaction (PCR), to determine the presence of the target bacterium and compare the results with the data obtained by the robot. Both laboratory and field tests were done. The highest match was obtained using Partial Least Squares-Discriminant Analysis PLS-DA, with a 66.4% detection rate for images obtained in the laboratory and 59.8% for images obtained in the field.

Background and Aims Bunch compactness is a key feature determining grape and wine composition because tight bunches show a less homogeneous ripening, and are prone to greater fungal disease incidence. The Organisation Internationale de la Vigne et du Vin descriptor, the most recent method for the assessment of bunch compactness, requires visual inspection and trained evaluators, and provides subjective and qualitative values. The aim of this work was to develop a methodology based on image analysis to determine bunch compactness in a non-invasive, objective and quantitative way. Methods and Results Ninety bunches of nine different red cultivars of Vitis vinifera L. were photographed with a colour camera, and their bunch compactness was determined by visual inspection. A predictive partial least squares (PLS) model was developed in order to estimate bunch compactness from the morphological features extracted by automated image analysis, after the supervised segmentation of the images. The PLS model showed a capability of 85.3% for predicting correctly the rating of bunch compactness. The most discriminant variables of the model were highly correlated with the tightness of the berries in the bunch (proportion of visibility of berries, rachis and holes) and with the shape of the bunch (roundness, compactness shape factor and aspect ratio). Conclusions The non-invasive, image analysis methodology presented here enables the quantitative assessment of bunch compactness, thereby providing precise objective information for this key parameter. Significance of the Study A quantitative, objective and accurate system based on image analysis was developed as an alternative to current visual methods for the estimation of bunch compactness. This novel method could be applied to the classification of table grapes and/or at the receival point of wineries for sorting and assessment of wine grapes before vinification.

[EN] This article presents a new method for analysing mosaics based on the mathematical principles of Symmetry Groups. This method has been developed to get the understanding present in patterns by extracting the objects that form them, their lattice, and the Wallpaper Group. The main novelty of this method resides in the creation of a higher level of knowledge based on objects, which makes it possible to classify the objects, to extract their main features (Point Group, principal axes, etc.), and the relationships between them. In order to validate the method, several tests were carried out on a set of Islamic Geometric Patterns from different sources, for which the Wallpaper Group has been successfully obtained in 85% of the cases. This method can be applied to any kind of pattern that presents a Wallpaper Group. Possible applications of this computational method include pattern classification, cataloguing of ceramic coatings, creating databases of decorative patterns, creating pattern designs, pattern comparison between different cultures, tile cataloguing, and so on., The authors wish to thank the Patronato de la Alhambra y Generalife (Granada, Spain) and the Patronato del Real Alcázar de Sevilla (Seville, Spain) for their valuable collaboration in this research work.

Individual items of any agricultural commodity are different from each other in terms of colour, shape or size. Furthermore, as they are living thing, they change their quality attributes over time, thereby making the development of accurate automatic inspection machines a challenging task. Machine vision-based systems and new optical technologies make it feasible to create non-destructive control and monitoring tools for quality assessment to ensure adequate accomplishment of food standards. Such systems are much faster than any manual non-destructive examination of fruit and vegetable quality, thus allowing the whole production to be inspected with objective and repeatable criteria. Moreover, current technology makes it possible to inspect the fruit in spectral ranges beyond the sensibility of the human eye, for instance in the ultraviolet and near-infrared regions. Machine vision-based applications require the use of multiple technologies and knowledge, ranging from those related to image acquisition (illumination, cameras, etc.) to the development of algorithms for spectral image analysis. Machine vision-based systems for inspecting fruit and vegetables are targeted towards different purposes, from in-line sorting into commercial categories to the detection of contaminants or the distribution of specific chemical compounds on the product’s surface. This chapter summarises the current state of the art in these techniques, starting with systems based on colour images for the inspection of conventional colour, shape or external defects and then goes on to consider recent developments in spectral image analysis for internal quality assessment or contaminant detection.

[EN] The feasibility of using visible and near-infrared spectroscopy technology combined with multivariate analysis to discriminate cv. 'Big Top' and cv. 'Diamond Ray' nectarines has been studied. These varieties are very difficult to differentiate visually on the production line but show important differences in taste that affects the acceptance by final consumers. The relationship between the diffuse reflectance spectra and the two nectarine varieties was established. Five hundred nectarine samples (250 of each variety) were used for the study. Tests were performed by using a spectrometer capable of measuring in two different spectral ranges (600-1100 nm and 900-1700 nm). These spectral ranges were used to develop two accurate classification models based on linear discriminate analysis (LDA) and partial least squares discriminate analysis (PLS-DA). Later, selection techniques were applied to select the most effective wavelengths. The results showed that the PLS-DA model achieved better accuracy and less latent variables than LDA model, and specifically, good results with 100% classification accuracy were obtained using only the 600-1100 nm spectral range for the two models and eight selected wavelengths. These results places visible and near-infrared spectrocopy as an accurate classification tool for nectarine varieties with a very similar appearance but different tastes that could be potentially used in an automated inspection system., This work was partially funded by the Generalitat Valenciana through the project AICO/2015/122 and by the INIA and FEDER funds through projects RTA2012-00062-C04-03 and RTA2015-00078-00-00. Victoria Cortes Lopez thanks the Spanish Ministry of Education, Culture and Sports for the FPU grant (FPU13/04202). The authors wish to thank the cooperative 'Fruits de Ponent' for providing the fruit.

The internal quality of intact persimmon cv. Rojo Brillante was assessed trough visible and near infrared hyperspectral imaging. Fruits at three stages of commercial maturity were exposed to different treatments with CO2 to obtain fruit with different ripeness and level of astringency (soluble tannin content). Spectral and spatial information were used for building classification models to predict ripeness and astringency trough multivariate analysis techniques like linear and quadratic discriminant analysis (LDA and QDA) and support vector machine (SVM). Additionally, flesh firmness was predicted by partial least square regression (PLSR). The full spectrum was used to determine the internal properties and later principal component analysis (PCA) was used to select optimal wavelengths (580, 680 and 1050 nm). The correct classification was above 92% for the three classifiers in the case of ripeness and 95% for QDA in the case of astringency. A value of R2 = 0.80 and a ratio of prediction deviation (RPD) of 1.86 were obtained with the selected wavelengths for the prediction of firmness which demonstrated the potential of hyperspectral imaging as a non-destructive tool in the assessment of the firmness, ripeness state and astringency level of Rojo Brillante persimmon., This work has been partially funded by the INIA and FEDER through projects RTA2012-00062-C04-01, RTA2012-00062-C04-03 and RTA2013-00043-C02, GVA through the project AICO/2015/122, the International S&T Cooperation Programs of China (2015DFA71150), and the International S&T Cooperation Program of Guangdong Province, China (2013B051000010). Sandra Munera thanks INIA for the grant FPI-INIA #43 (CPR2014-0082) partially supported by FSE funds.

[EN] Development of non-destructive tools for determining mango ripeness would improve the quality of industrial production of the postharvest processes. This study addresses the creation of a new sensor that combines the capability of obtaining mechanical and optical properties of the fruit simultaneously. It has been integrated into a robot gripper that can handle the fruit obtaining non-destructive measurements of firmness, incorporating two spectrometer probes to simultaneously obtain reflectance properties in the visible and near-infrared, and two accelerometers attached to the rear side of two fingers. Partial least square regression was applied to different combinations of the spectral data obtained from the different sensors to determine the combination that provides the best results. Best prediction of ripening index was achieved using both spectral measurements and two finger accelerometer signals, with R2 P ¿ 0:832 and RMSEP of 0.520. These results demonstrate that simultaneous measurement and analysis of the data fusion set improve the robot gripper features, allowing assessment of the quality of the mangoes during pick and place operations., This work has been partially funded by the Instituto Nacional de Investigacion y Tecnologia Agraria y Alimentaria de Espana (INIA) and FEDER through research projects RTA2012-00062-C04-01, RTA2012-00062-C04-02, RTA2012-00062-C04-03, and by the Conselleria d' Educacio, Investigacio, Cultura i Esport, Generalitat Valenciana, through the project AICO/2015/122. V. Cortes thanks the Spanish MEC for the FPU grant (FPU13/04202).

BACKGROUND: Berry weight, berry number and cluster weight are key parameters for yield estimation for wine and tablegrape industry. Current yield prediction methods are destructive, labour-demanding and time-consuming. In this work, a new methodology, based on image analysis was developed to determine cluster yield components in a fast and inexpensive way. RESULTS: Clusters of seven different red varieties of grapevine (Vitis vinifera L.) were photographed under laboratory conditions and their cluster yield components manually determined after image acquisition. Two algorithms based on the Canny and the logarithmic image processing approaches were tested to find the contours of the berries in the images prior to berry detection performed by means of the Hough Transform. Results were obtained in two ways: by analysing either a single image of the cluster or using four images per cluster from different orientations. The best results (R 2 between 69% and 95% in berry detection and between 65% and 97% in cluster weight estimation) were achieved using four images and the Canny algorithm. The model’s capability based on image analysis to predict berry weight was 84%. CONCLUSION: The new and low-cost methodology presented here enabled the assessment of cluster yield components, saving time and providing inexpensive information in comparison with current manual methods. © 2014 Society of Chemical Industry

The mechanisation and automation of citrus harvesting is considered to be one of the best options to reduce production costs. Computer vision technology has been shown to be a useful tool for fresh fruit and vegetable inspection, and is currently used in post-harvest fruit and vegetable automated grading systems in packing houses. Although computer vision technology has been used in some harvesting robots, it is not commonly utilised in fruit grading during harvesting due to the difficulties involved in adapting it to field conditions. Carrying out fruit inspection before arrival at the packing lines could offer many advantages, such as having an accurate fruit assessment in order to decide among different fruit treatments or savings in the cost of transport and marketing non-commercial fruit. This work presents a computer vision system, mounted on a mobile platform where workers place the harvested fruits, that was specially designed for sorting fruit in the field. Due to the specific field conditions, an efficient and robust lighting system, very low-power image acquisition and processing hardware, and a reduced inspection chamber had to be developed. The equipment is capable of analysing fruit colour and size at a speed of eight fruits per second. The algorithms developed achieved prediction accuracy with an R-2 coefficient of 0.993 for size estimation and an R-2 coefficient of 0.918 for the colour index., This research work has been funded by the Instituto Nacional de Investigacion y Tecnologia Agraria y Alimentaria de Espana (INIA) and the European FEDER funds (projects RTA2009-00118-C02-01 and RTA2009-00118-C02-02). The authors wish to thank the collaboration of the company Argiles Diseny i Fabricacio, S.L.

[EN] This work reports the development of automated systems based on computer vision to improve the quality control and sorting of dried figs of Cosenza (protected denomination of origin) focusing on two research issues. The first was based on qualitative discrimination of figs through colour assessment comparing the analysis of colour images obtained using a digital camera with those obtained according to conventional instrumental methods, i.e. colourimetry currently done in laboratories. Data were expressed in terms of CIE XYZ, CIELAB and HunterLab colour spaces, as well as the browning index measurement of each fruit, and then, analysed using PCA and PLS-DA based methods. The results showed that both chroma meter and image analysis allowed a complete distinction between high quality and deteriorated figs, according to colour attributes. The second research issue had the purpose of developing image processing algorithms to achieve real-time sorting of figs using an experimental prototype based on machine vision, simulating an industrial application. An extremely high 99.5% of deteriorated figs were classified correctly as well as 89.0% of light coloured good quality figs A lower percentage was obtained for dark good quality figs but results were acceptable since the most of the confusion was among the two classes of good product. (c) 2015 Elsevier B.V. All rights reserved., This work has been partially funded by INIA through research project RTA2012-00062-C04-01 with the support of European FEDER funds.

The quality standards of citrus for fresh consumption are mainly based on the absence of bruises and rotting, as well as an adequate shape, color, and size. Current industrial systems for inspection require the processing of images in a few milliseconds, inspecting the whole surface of the fruit moving on a conveyor belt. Most of them are limited to the visible region of the electromagnetic spectrum, but the identification of defects is an important task that still remains unresolved. Information provided by near infrared or ultraviolet regions of the spectrum may improve the accuracy of fruit identification and the detection of dangerous damages or diseases that evolve or are invisible to our eyes. Furthermore, some techniques open the possibility of inspecting the internal parts of the fruits. This chapter summarizes research related to the application of machine vision, including different imaging technologies and techniques for the automated inspection of the quality of citrus fruits.

[EN] In this work an N-way partial least squares regression discriminant analysis (NPLS-DA) methodology is developed to detect symptoms of disease caused by Penicillium digitatum in citrus fruits (green mould) using visible/near infrared (VIS/NIR) hyperspectral images. To build the discriminant model a set of oranges and mandarins was infected by the fungus and another set was infiltrated just with water for control purposes. A double cross-validation strategy is used to validate the discriminant models. Finally, permutation testing is used to select a few bands offering the best correct classification rates in the validation set. The discriminant models developed here can be potentially implemented in a fruit packinghouse to detect infected citrus fruits at their arrival from the field with affordable multispectral (3 5 channels) cameras installed in the packinglines., This research was partially funded by the Spanish Ministry of Science and Innovation through grants DPI2011-28112-C04-02 and DPI2014-55276-C05-1R, and by INIA through grant RTA2012-00062-C04-01. In all cases with the support of European FEDER funds. Authors thank Lluis Palou from the Centro de Tecnologia Postcosecha at the IVIA for the help and supervision in the innoculation process of the fruits.

This study was conducted to develop a fast and accurate computer-based machine vision system for detecting irregular potatoes in real-time. Supported algorithms were specifically developed and programmed for image acquisition and processing, controlling the whole process, saving the classification results and monitoring the progress of all operations. A database of images was first formulated from potatoes with different shapes and sizes, and then some essential geometrical features such as perimeter, centroid, area, moment of inertia, length and width were extracted from each image. Also, eight shape parameters originated from size features and Fourier transform were calculated for each image in the database. All extracted shape parameters were entered in a stepwise linear discriminant analysis to extract the most important parameters that most characterized the regularity of potatoes. Based on stepwise linear discriminant analysis, two shape features (roundness and extent) and four Fourier-shape descriptors were found to be effective in sorting regular and irregular potatoes. The average correct classification was 96.5% for a training set composed of 228 potatoes and then the algorithm was validated in another testing set composed of 182 potatoes in a real-time operation. The experiments showed that the success of in-line classification of moving potatoes was 96.2%. Concurrently, the well-shaped potatoes were classified by size achieving a 100% accuracy indicating that the developed machine vision system has a great potential in automatic detection and sorting of misshapen products. (c) 2012 Elsevier Ltd. All rights reserved.

Penicillium fungi are among the main defects that may affect the commercialization of citrus fruits. Economic losses in fruit production may become enormous if an early detection of that kind of fungi is not carried out. That early detection is usually based either on UltraViolet light carried out manually. This work presents a new approach based on hyperspectral imagery for defect segmentation. Both the physical device and the data processing (geometric corrections and band selection) are presented. Achieved results using classifiers based on Artificial Neural Networks and Decision Trees show an accuracy around 98%; it shows up the suitability of the proposed approach.

Early automatic detection of fungal infections in post-harvest citrus fruits is especially important for the citrus industry because only a few infected fruits can spread the infection to a whole batch during operations such as storage or exportation, thus causing great economic losses. Nowadays, this detection is carried out manually by trained workers illuminating the fruit with dangerous ultraviolet lighting. The use of hyperspectral imaging systems makes it possible to advance in the development of systems capable of carrying out this detection process automatically. However, these systems present the disadvantage of generating a huge amount of data, which must be selected in order to achieve a result that is useful to the sector. This work proposes a methodology to select features in multi-class classification problems using the receiver operating characteristic curve, in order to detect rottenness in citrus fruits by means of hyperspectral images. The classifier used is a multilayer perceptron, trained with a new learning algorithm called extreme learning machine. The results are obtained using images of mandarins with the pixels labelled in five different classes: two kinds of sound skin, two kinds of decay and scars. This method yields a reduced set of features and a classification success rate of around 89%., This work was partially funded by the Instituto Nacional de Investigacion y Tecnologia Agraria y Alimentaria de Espana (INIA) through research project RTA2009-00118-C02-01 and by the Ministerio de Ciencia e Innovacion de Espana (MICINN) through research project DPI2010-19457, both projects with the support of European FEDER funds. This work was also been partially funded by Universitat de Valencia through project UV-INVAE11-41271.

The best alternative for reducing citrus production costs is mechanization. Machine vision is a reliable technology for the automatic inspection of fresh fruits and vegetables that can be adapted to harvesting machines. In these, fruits can be inspected before sending them to the packinghouse and machine vision provides important information for subsequent processing and avoids spending further resources in non-marketable fruit. The present work describes a computer vision system installed on a harvesting machine developed jointly by IVIA and a Spanish enterprise. In this machine, hand pickers directly drop the fruit as they collect it, which results in an important increase of productivity. The machine vision system is placed over rollers in order to inspect the produce, and separate those that can be directly sent to the fresh market from those that do not meet minimal quality requirements but can be used by the processing industry, based on color, size and the presence of surface damages. The system was tested under field conditions.

The computer vision systems currently used for the automatic inspection of citrus fruits are normally based on supervised methods that are capable of detecting defects on the surface of the fruit but are unable to discriminate between different types of defects. identifying the type of the defect affecting each fruit is very important in order to optimise the marketing profit and to be able to take measures to prevent such defects from occurring in the future. In this paper, we present a computer vision system that was developed for the recognition and classification of the most common external defects in citrus. in order to discriminate between 11 types of defects, images of the defects were acquired in five spectral areas, including the study of near infrared reflectance and ultraviolet induced fluorescence. The system combines spectral information about the defects with morphological estimations of them in order to classify the fruits in categories. The fruit-sorting algorithm proposed here was tested by using it to identify the defects in more than 2000 citrus fruits, including mandarins and oranges. The overall success rate reached 86%.

BACKGROUND: The sterile insect technique (SIT) is acknowledged around the world as an effective method for biological pest control of Ceratitis capitata (Wiedemann). Sterile insects are produced in biofactories where one key issue is the selection of the progenitors that have to transmit specific genetic characteristics. Recombinant individuals must be removed as this colony is renewed. Nowadays, this task is performed manually, in a process that is extremely slow, painstaking and labour intensive, in which the sex of individuals must be identified. The paper explores the possibility of using vision sensors and pattern recognition algorithms for automated detection of recombinants. RESULTS: An automatic system is proposed and tested to inspect individual specimens of C. capitata using machine vision. It includes a backlighting system and image processing algorithms for determining the sex of live flies in five high-resolution images of each insect. The system is capable of identifying the sex of the flies by means of a program that analyses the contour of the abdomen, using fast Fourier transform features, to detect the presence of the ovipositor. Moreover, it can find the characteristic spatulate setae of males. Simulation tests with 1000 insects (5000 images) had 100% success in identifying male flies, with an error rate of 0.6% for female flies. CONCLUSION: This work establishes the basis for building a machine for the automatic detection and removal of recombinant individuals in the selection of progenitors for biofactories, which would have huge benefits for SIT around the globe. Copyright © 2008 Society of Chemical Industry

The presence of skin defects is one of the most influential factors in the price of fruit. The detection of defects during packing operations ensures that only fruits with a good quality reach the market. Moreover, the identification of the type of each defect will increase both the quality of the fruit and also the producer's profit. At the present time, fruit with slight defects is marketed together with sound fruit, thus depreciating the quality of the batch, or it is removed together with seriously damaged fruit, thereby causing economic losses. Most current computer vision systems used in the automatic quality inspection of food are limited to the visible region of the electromagnetic spectrum as they tend to imitate the human eye. However, non-visible information, such as that provided by near-infrared or ultraviolet regions of the spectrum, can improve the inspection by detecting specific defects or allowing the detection of non-visible damages. This work summarises our research in the application of near-infrared, ultraviolet and fluorescence computer vision systems in the identification of the most common defects of citrus fruits, and proposes a fruit sorting algorithm that combines this different spectral information (including visible) to classify fruit according to the type of defect. Results showed that the contribution of non-visible information can improve the detection and identification of some defects. Compared with the results from colour images, the detection accuracy of anthracnose increased from 86% by using NIR images; and the accuracy of green mould was increased from 65% to 94% by using images of fluorescence. (c) 2007 Elsevier Ltd. All rights reserved.

The vibratory behaviour of citrus fruits is studied using slow-motion cameras in order to gain a better understanding of the parameters involved in fruit detachment when mechanical harvesting is done using shakers. Single citrus fruits with a small portion of stem were vibrated using strokes from 60 mm to 180 mm and frequencies from 3 Hz to 18 Hz. The movement was recorded at 300 fps and the main parameters considered for fruit detachment were determined through the analysis of the video sequences. Image-processing algorithms created for this purpose were applied to the automated estimation of the centroid of the fruit, the angle of the stem pistil axis, and the position of some selected points in the fruit in each frame of the video sequences to obtain dynamic parameters such as the position, speed and acceleration of the fruit during the movement until it is detached. The signals obtained from the image processing were filtered, providing results in accordance with the calibration systems. In general, results suggest that the inertial forces transmitted to the fruit were lower than the tensile forces required to detach the fruit by pulling it in the stem pistil direction. The largest peaks were observed using long strokes that required fewer cycles for detachment. On the other hand, short strokes combined with high frequencies needed more cycles, and thus a fatigue phenomenon was present. Short strokes and low frequencies were unable to detach some fruit. (C) 2014 IAgrE. Published by Elsevier Ltd. All rights reserved., This work was founded by the Instituto Nacional de Investigacion y Tecnologia Agraria y Alimentaria (INIA) through the projects RTA2009-00118-C02-01 and RTA2009-00118-C02-02, and co-founded by European FEDER founds.

Hyperspectral systems are characterised by offering the possibility of acquiring a large number of images at different consecutive wavebands. To ensure reliable and repeatable results using this kind of optical sensors, the intensity shown by the objects in the different spectral images must be independent from the differences in sensitivity of the system for the different wavelengths. The spectral efficiency of the acquisition devices and the spectral emission of the lighting system vary across the spectrum and the images, and therefore the results can reproduce these variations if the system is not properly calibrated and corrected. This is particularly complex, when several LCTF devices are used to obtain large spectral ranges. This work presents the development of a hyperspectral system based on two liquid crystal tuneable filters for the acquisition of images of spherical fruits. It also proposes a methodology for acquiring and segmenting images of citrus fruits aimed at detecting decay in citrus fruits that has been capable of correctly classifying 98 % of pixels as rotten or non-rotten and 95 % of fruit., This work has been partially funded by the Instituto Nacional de Investigacion y Tecnologia Agraria y Alimentaria de Espana (INIA) through research project RTA2012-00062-C04-01 and RTA2012-00062-C04-03 with the support of European FEDER funds, the Universitat de Valencia through project UV-INV-AE11-41271, and the UPV-IVIA through collaboration agreement UPV-2013000005.

Mango fruit are sensitive and can easily develop brown spots after suffering mechanical stress during postharvest handling, transport and marketing. The manual inspection of this fruit used today cannot detect the damage in very early stages of maturity and to date no automatic tool capable of such detection has been developed, since current systems based on machine vision only detect very visible damage. The application of hyperspectral imaging to the postharvest quality inspection of fruit is relatively recent and research is still underway to find a method of estimating internal properties or detecting invisible damage. This work describes a new system to evaluate mechanically induced damage in the pericarp of ‘Manila’ mangos at different stages of ripeness based on the analysis of hyperspectral images. Images of damaged and intact areas of mangos were acquired in the range 650–1100 nm using a hyperspectral computer vision system and then analysed to select the most discriminating wavelengths for distinguishing and classifying the two zones. Eleven feature-selection methods were used and compared to determine the wavelengths, while another five classification methods were used to segment the resulting multispectral images and classify the skin of the mangos as sound or damaged. A 97.9% rate of correct classification of pixels was achieved on the third day after the damage had been caused using k-Nearest Neighbours and the whole spectra and the figure dropped to 91.4% when only the most discriminant bands were used.

A key aspect for the consumer when it comes to deciding on a particular product is the colour. In order to make fruit available to consumers as early as possible, the collection of oranges and mandarins begins before they ripen fully and reach their typical orange colour. As a result, they are therefore subjected to certain degreening treatments, depending on their standard colour citrus index at harvest. Recently, a mobile platform that incorporates a computer vision system capable of pre-sorting the fruit while it is being harvested has been developed as an aid in the harvesting task. However, due to the restrictions of working in the field, the computer vision system developed for this machine is limited in its technology and processing capacity compared to conventional systems. This work shows the optimised algorithms for estimating the colour of citrus in-line that were developed for this mobile platform and its performance is evaluated against that of a spectrophotometer used as a reference in the analysis of colour in food. The results obtained prove that our analysis system predicts the colour index of citrus with a good reliability (R2 = 0.925) working in real time. Findings also show that it is effective for classifying harvested fruits in the field according to their colour. © 2012 Springer Science+Business Media New York., This work was partially funded by the INIA through research project RTA2009-00118-C02-01 with the support of European FEDER funds, and by the project PAID-05-11-2745, Vicerectorat d'Investigacio, Universitat Politecnica de Valencia.

Citrus are one of the major fruit production in Spain. Most of this production is exported to Europe for fresh consumption, where consumers increasingly demand an extra quality. Application of machine vision is a priority, in order to compete with third countries that have much lower production costs. Current commercial sorters based on machine vision only solve the problems that require less computing time, as for instance, sizing or classification in colours, or work with low resolution images, in order to achieve high processing speeds. However, this approach reduces the accuracy of the system when estimating the size of the fruit. Another important fact that needs consideration is the possibility of detecting defects on the skin surface using wavelengths that are outside the visible spectrum. This paper describes the work carried out in order to develop a machine vision system capable to inspect citrus at a rate of 10 fruits per second. The work involved the development of a multispectral camera, parallel algorithms and their implementation on a specific board based on two DSP that work in parallel.

Processed olives have traditionally been classified by human experts but the process is slow, expensive and without a good repeatability. The main objective of this paper is to design a system to sort olives automatically using machine vision. First of all, human experts classified different batches of olives in four categories to extract the parameters related with each class. After characterising the processed olives, an algorithm to sort automatically olives was implemented and tested. Different resolutions were considered to study the effect of resolution in the classification results. Finally, a comparison between human and machine results is done.

Current grading lines for fresh fruit use sensors that measure weight, size, sometimes colour or firmness. However, none of them measures so far an important organoleptic criteria for the consumer: sugar content. Image analysis can provide colour information on the fruit, while near infrared spectrophotometric data can be used in order to determine sugar content. Sensor fusion methodology is aimed at improving the sugar content prediction by combining both on-line non-destructive sensors. The fusion process began by a diagnosis of each sensor, followed by a neural network modelisation that uses the information provided by the sensors. It was found that the repeatability of the classification of the fruits based on sugar content was improved when two sensors were combined. The sensors and the fusion process were implemented on-line within a robotics device currently running at 3.5 seconds per fruit., Les lignes de tri actuelles destinées à l'évaluation de la qualité des fruits frais utilisent des capteurs qui mesurent le poids, la taille, et parfois la couleur ou la fermeté des fruits. Toutefois, aucune de ces lignes n'est équipée d'un système de mesure du paramètre organoleptique le plus important pour le consommateur: la teneur en sucres. L'analyse d'image peut fournir une information colorimétrique du fruit, tandis que la spectrophotométrie peut être utilisée pour déterminer la teneur en sucres de manière non-destructive. La fusion de capteurs vise à améliorer la prédiction non destructive de la teneur en sucres, en combinant ces deux capteurs en ligne. La fusion s'opère en deux temps; dans un premier temps, un diagnostic est réalisé sur l'état de la mesure fournie par chaque capteur, puis un réseau de neurones combine les informations fournies par chacun des capteurs. Les résultats démontrent une plus grande répétabilité de la classification des fruits à partir de leur teneur en sucres, lorsque la fusion de capteurs est mise en oeuvre. Les capteurs et le système de fusion ont été installés en ligne au sein d'un système robotisé fonctionnant actuellement à environ 3.5 s par fruit.

[EN] Green mold (Penicillium digitatum) and blue mold (Penicillium italicum) are important sources of postharvest decay affecting the commercialization of mandarins. These fungi infections produce enormous economic losses in mandarin production if early detection is not carried out. Nowadays, this detection is performed manually in dark chambers, where the fruit is illuminated by ultraviolet light to produce fluorescence, which is potentially dangerous for humans. This paper documents a new methodology based on hyperspectral imaging and advanced machine-learning techniques (artificial neural networks and classification and regression trees) for the segmentation and classification of images of citrus free of damage and affected by green mold and blue mold. Feature selection methods are used in order to reduce the dimensionality of the hyperspectral images and determine the 10 most relevant. Neural Networks were used to segment the hyperspectral images. Result's achieved using classifiers based on decision trees show an accuracy of around 93% in the problem of decay classification., This work was partially funded by the University of Valencia through project UV-INV-AE11-41271 and by the Instituto Nacional de Investigacion y Tecnologia Agraria y Alimentaria de Espana (INIA) through research projects RTA2012-00062-C04-01 and RTA2012-00062-C04-03

Hyperspectral imaging systems allow to detect the initial stages of decay caused by fungi in citrus fruit automatically, instead of doing it manually under dangerous ultraviolet illumination, thus preventing the fungal infestation of other sound fruit and, consequently, the enormous economical losses generated. However, these systems present the disadvantage of generating a huge amount of data, which is necessary to select for achieving some result useful for the sector. There are numerous feature selection methods to reduce dimensionality of hyperspectral images. This work compares a feature selection method using the area under the receiver operating characteristic (ROC) curve with other common feature selection techniques, in order to select an optimal set of wavelengths effective in the detection of decay in a citrus fruit using hyperspectral images. This comparative study is done using images of mandarins with the pixels labelled in five different classes: two types of healthy skin, two types of decay and scars, ensuring that the ROC technique generally provides better results than the other methods. © 2012 Springer Science+Business Media, LLC., This work has been partially funded by the Universitat de València through project UV-INV-AE11-41271, by the Instituto Nacional de Investigación y Tecnología Agraria y Alimentaria de España (INIA) through research project RTA2009-00118-C02-01 and by the Ministerio de Ciencia e Innovación de España (MICINN) through research project DPI2010-19457, both projects with the support of European FEDER funds.

The citrus industry is nowadays an important part of the Spanish agricultural sector. One of the main problems present in the citrus industry is decay caused by Penicillium digitatum and Penicillium italicum fungi. Early detection of decay produced by fungi in citrus is especially important for the citrus industry of distribution. This chapter presents a hyperspectral computer vision system and a set of machine learning techniques in order to detect decay caused by Penicillium digitatum and Penicillium italicum fungi that produce more economic losses to the sector. More specifically, the authors employ a hyperspectral system and artificial neural networks. Nowadays, inspection and removal of damaged citrus is done manually by workers using dangerous ultraviolet light. The proposed system constitutes a feasible and implementable solution for the citrus industry; this has been proven by the fact that several machinery enterprises have shown their interest in the implementation and patent of the system.

Nowadays, there is a growing demand for quality fruits and vegetables that are simple to prepare and consume, like minimally processed fruits. These products have to accomplish some particular characteristics to make them more attractive to the consumers, like a similar appearance and the total absence of external defects. Although recent advances in machine vision have allowed for the automatic inspection of fresh fruit and vegetables, there are no commercially available equipments for sorting of minority processed fruits, like arils of pomegranate (Punica granatum L) or segments of Satsuma mandarin (Citrus unshiu) ready to eat. This work describes a complete solution based on machine vision for the automatic inspection and classification of these fruits based on their estimated quality. The classification is based on morphological and colour features estimated from images taken in-line, and their analysis using statistical methods in order to grade the fruit into commercial categories.

Machine vision has become a key technology in the area of quality control of horticultural products due to the increasing quality demands of manufacturers and customers, providing a fast and reliable method of inspecting features such as ripeness, size, weight, shape, colour, the presence of blemishes and diseases, the presence or absence of fruit stems, the presence of seeds, sugar content and so forth. This chapter provides a comprehensive review of recent advances in computer vision applied to the inspection of nuts, fruits and vegetables and outlines potential applications in the industry.

Publisher Summary Some of the most important aspects that need to be taken into consideration when developing a hyperspectral inspection system for citrus include the geometry of the fruit, the emission spectrum of the lighting source, and their interaction. Because many citrus fruits are almost spherical, each point of their surface reflects the electromagnetic radiation differently toward the camera. This causes a gradual darkening of the image especially the further pixels from the light source, which is a phenomenon that must be artificially corrected. In addition, the variation of the efficiency of the filters with the wavelength should be also taken into consideration to enable the appropriate corrections to obtain true reflectance images. Hyperspectral systems are an important tool for the quality inspection of citrus fruits, offering the possibility of designing machines for the automatic identification of blemishes. This is particularly important for early rot detection, one of the major problems faced by this sector. However, a realistic implementation of such systems probably still requires an important effort in adequately reducing the number of input bands.

The Bayesian approach to classification is intended to solve questions concerning how to assign a class to an observed pattern using probability estimations. Red, green and blue (RGB) or hue, saturation and lightness (HSL) values of pixels in digital colour images can be considered as feature vectors to be classified, thus leading to Bayesian colour image segmentation. Bayesian classifiers are also used to sort objects but, in this case, reduction of the dimensionality of the feature vector is often required prior to the analysis. This chapter shows some applications of Bayesian learning techniques in computer vision in the agriculture and agri-food sectors. Inspection and classification of fruit and vegetables, robotics, insect identification and process automation are some of the examples shown. Problems related with the natural variability of colour, sizes and shapes of biological products, and natural illuminants are also discussed. Moreover, implementations that lead to real-time implementation are explained.

One of the main problems in the post-harvest processing of citrus is the detection of visual defects in order to classify the fruit depending on their appearance. Species and cultivars of citrus present a high rate of unpredictability in texture and colour that makes it difficult to develop a general, unsupervised method able of perform this task. In this paper we study the use of a general approach that was originally developed for the detection of defects in random colour textures. It is based on a Multivariate Image Analysis strategy and uses Principal Component Analysis to extract a reference eigenspace from a matrix built by unfolding colour and spatial data from samples of defect-free peel. Test images are also unfolded and projected onto the reference eigenspace and the result is a score matrix which is used to compute defective maps based on the T2 statistic. In addition, a multiresolution scheme is introduced in the original method to speed up the process. Unlike the techniques commonly used for the detection of defects in fruits, this is an unsupervised method that only needs a few samples to be trained. It is also a simple approach that is suitable for real-time compliance. Experimental work was performed on 120 samples of oranges and mandarins from four different cultivars: Clemenules, Marisol, Fortune, and Valencia. The success ratio for the detection of individual defects was 91.5%, while the classification ratio of damaged/sound samples was 94.2%. These results show that the studied method can be suitable for the task of citrus inspection. © 2010 Elsevier B.V. All rights reserved., This work has been supported by the Spanish Ministry of Education (MEC) and by European FEDER funds, through the research projects DPI2007-66596-C02-01 (VISTAC) and DPI-2007-66596-C02-02.

The pomegranate is a fruit with excellent organoleptic and nutritional properties, but the fact that it is difficult to peel affects its commercialisation and decreases its potential consumption. One solution is to market the arils of pomegranate in a ready-to-eat form. However, after the peeling process, unwanted material, such as internal membranes and defective arils, is extracted together with good arils and must be removed on the packing line because the presence of such material shortens the shelf life of the product or deteriorates its appearance. For different reasons, the commercial sorting machines that are currently available for similar commodities (cherries, nuts, rice, etc.) are not capable of handling and sorting pomegranate arils, thus making it necessary to build specific equipment. This work describes the development of a computer vision-based machine to inspect the raw material coming from the extraction process and classify it in four categories. The machine is capable of detecting and removing unwanted material and sorting the arils by colour. The prototype is composed of three units, which are designed to singulate the objects to allow them be inspected individually and sorted. The inspection unit relies on a computer vision system. Two image segmentation methods were tested: one uses a threshold on the R/G ratio and the other is a more complex approach based on Bayesian Linear Discriminant Analysis (LDA) in the RGB space. Both methods offered an average success rate of 90% on a validation set, the former being more intuitive for the operators, as well as faster and easier to implement, and for these reasons it was included in the prototype. Subsequently, the complete machine was tested in industry by working in real conditions throughout a whole pomegranate season, in which it automatically sorted more than nine tons of arils.

The detection of rotten citrus in packing lines is carried out manually under ultraviolet illumination, which is dangerous for workers. Light emitted by the rotten region of the fruit due to the ultraviolet-induced fluorescence is used by the operator to detect the damages. This procedure is required because the low contrast between the damaged and sound skin under visible illumination difficult their detection. We study a set of techniques aimed to detect rottenness in citrususing visible and near infrared lighting trough an hyperspectral imaging system. Methods for selecting a proper set of wavelengths are investigated such as correlation analysis, mutual information, stepwise or genetic algorithms. The image segmentation relies on the combination of band selection techniques and pixel classification methods such as classification and regression trees and linear discriminant analysis.

This study proposes a method for correcting the adverse effects produced by the curvature of spherical objects in acquiring images with a computer vision system. Its suitability has been illustrated in a specific case of citrus fruits. The images of this kind of fruit are darker in areas nearer the edge than in the centre, and this makes them more difficult to analyse. This methodology considers the fruit as being a Lambertian ellipsoidal surface and produces a 3D model of the fruit. By doing it becomes possible to calculate the part of the radiation that should really reach the camera and to make the intensity of the radiation uniform over the whole of the fruit surface captured by the camera, no matter what region is being sampled. Some tests have been carried out in order to prove that using the proposed correction methodology the reflectance in all the surface of the fruit is similar, minimising the differences from the central area to the peripheral areas. The methodology presented here has been tested using a hyperspectral computer vision system based on tunable liquid crystal filters and it has proved to be effective for minimising the adverse side effects produced by the curvature of the fruit on the intensity of the radiation captured by the camera. The experiments show that applying our method homogenises the grey level of the pixels belonging to the same class, regardless of the region of the fruit surface they are from; standard deviation is also reduced, which facilitates subsequent classification tasks. (c) 2007 Elsevier Ltd. All rights reserved.

Nowadays, the detection of fruit infected with Penicillium sp. fungi on packing lines is carried out manually under ultraviolet illumination. Ultraviolet sources induce visible fluorescence of essential oils, present in the skin of citrus and which are released by the action of fungi, thus increasing the contrast between sound and rotten skin. This work analyses a set of techniques aimed at detecting rotten citrus without the use of UV lighting. The techniques used include hyperspectral image acquisition, pre-processing and calibration, feature selection and segmentation using linear and non-linear methods for classification of fruits. Different methods such as correlation analysis, mutual information, stepwise, and genetic algorithms based on linear discriminant analysis (LDA) are studied to select the most relevant bands. Image segmentation relies on the combination of efficient band selection techniques and also on pixel classification methods such as classification and regression trees (CART) and LDA. The results were obtained using a large dataset of images of mandarins cv. “Clemenules” by applying the CART method. The hyperspectral computer vision system proposed here is capable of detecting damage caused by Penicillium digitatum in mandarins using a reduced set of optimally selected bands.

This work focuses on the development of a computer vision system for the automatic on-line inspection and classification of Satsuma segments. During the image acquisition the segments are in movement, wet and frequently in contact with other pieces. The segments are transported over six semi-transparent conveyor belts that advance at speed of 1 m/s. During on-line operation, the system acquires images of the segments using two cameras connected to a single computer and process the images in less than 50 ms. Extracting morphological features from the objects, the system identifies automatically pieces of skin and row material and separates entire segments from broken ones, discriminating between those with slight or large breaking degree. Combinations of morphological parameters were employed to decide the quality of each segment, classifying correctly 95% of sound segments.

Presented is a full computer vision system for the identification of post-harvest damage in citrus packing houses. The method is based on the combined use of hyperspectral images and the Mahalanobis kernel classifier. More accurate and reliable results compared to other methods are obtained in several scenarios and acquired images.

Due to the high sorting speed required during fruit inspection and classification in packing lines, most of the current automatic systems, based on machine vision, normally employ supervised techniques oriented towards individual pixels to segment the images of the fruits. These techniques require previous training given by experts in order to classify the colour of each pixel as belonging to any of the regions of interest and frequent training sessions during normal operation throughout the season to adapt the system to the great colour variability present in biological products like fruits. In region-oriented segmentation algorithms, however, the contrast between different areas in the image becomes more important than the individual pixel colour, thus solving the problem related to the variability of the natural colour of fruits. This work proposes a region-oriented segmentation algorithm for detecting the most common peel defects of citrus fruits. Focused on the detection of the regions of interest consisting of the sound peel, the stem and the defects, this method is an original contribution that allows successful segmentation of smaller defects, such as scale. The algorithm was tested on images of different varieties of oranges and mandarins presenting defects, without any further training being given between inspections of different batches, even species, of citrus fruits. Assuming that the most surface of the fruit corresponds to sound peel, the proposed algorithm was able to correctly detect 95% of the defects under study. (c) 2006 Elsevier Ltd. All rights reserved.