Abstract: Tomatoes are widely cultivated, both by family farmers and corporate producers. During the tomato growth cycle, several diseases can affect the plant. The identification of these diseases through short-range images is significant, and computer vision techniques are commonly used to identify diseases in plant leaves. In this paper, a hybrid model that combines a convolutional neural network (CNN) and a Random Forest (RF) decision tree is used for foliar spot detection in tomato leaves. High-level features learned and extracted from CNN are used as input for the RF classifier. To evaluate the proposed model’s performance for plant disease identification, a case study of 2480 low-cost digital RGB images collected in actual field conditions, under different intensities of light exposure, were used, including healthy tomato leaves and leaves with visible symptoms of powdery mildew fungus, which attacks the tomato leaf. The results were compared with six conventional machine learning classifiers: Logistic Regression (LR), Linear Discriminant Analysis (LDA), K- Nearest Neighbors (KNN), Naive Bayes (NB), Support Vector Machine (SVM) and Random Forest (RF). The results show that the proposed model outperformed conventional classifiers, reaching an accuracy of 98%. The results highlight the importance of fusing models to improve the detection plant´s diseases.

ABSTRACT The constant monitoring of water quality is fundamental for the understanding of the aquatic environment, yet it demands great financial investments and is susceptible to inconsistencies and missing values. Using a database composed of 59 sampling campaigns, performed for 12 years, on 10 monitoring stations along the Iguassu River Basin (Southern Brazil), this study presents a model, based on feed-forward neural networks, which imputed 1,370 values for 11 traditional water quality parameters, as well as 3 contaminants of emerging concern (caffeine, estradiol and ethinylestradiol). The model validation errors varied from 0.978 mg L-1 and 0.017 mg L-1 for the traditional parameters, for caffeine the validation error was of 0.212 µg L-1 and for the hormones, the errors were of 0.04 µg L-1 (E1) and 0.044 µg L-1 (EE1). The models underwent two techniques to understand the operations performed within the model (isolation and nullification), which were consistent to those explained by natural processes. The results point to the validity of modeling water quality parameters (especially the concentrations of caffeine) through neural networks, which could lead to better resource allocation in environmental monitoring, as well as improving available datasets and valuing previous monitoring efforts.

RESUMO O monitoramento constante da qualidade da água é fundamental para o entendimento dos ambientes aquáticos, mas este esforço demanda grandes investimentos financeiros, além de estar suscetível a inconsistências e falhas na obtenção dos valores. Usando uma base de dados composta por 59 campanhas amostrais, realizadas durante um período de 12 anos, em 10 estações de monitoramento ao longo da bacia do rio Iguaçú, este trabalho apresentou um modelo baseado em redes neurais artificiais que preencheu 1.370 valores para onze parâmetros tradicionais de qualidade da água. Os erros de validação do modelo variaram de 0,978 mg L-1 a 0,017 mg L-1 para os parâmetros tradicionais, enquanto para a cafeína, este erro foi de 0,212 µg L-1 e para os hormônios de 0,04 µg L-1 (E1) e 0,044 µg L-1 (EE1). Duas técnicas (isolamento e anulação) foram aplicadas nos modelos para se entender o relacionamento treinado pelo modelo entre as variáveis de entrada e saída. Os resultados apontam para a viabilidade da aplicação de redes neurais para a modelagem de parâmetros de qualidade da água (em especial, a cafeína), o que poderia levar a uma melhor alocação de recursos no monitoramento ambiental, além de expandir os bancos de dados disponíveis e valorizar os esforços despendidos para este monitoramento.

RESUMO As alterações da qualidade da água são reflexo das dinâmicas de usos da água e do solo de uma bacia e é um desafio medi-las apropriadamente, especialmente quando se olha para variáveis bastante distintas, como os constituintes tradicionais da matéria orgânica e os compostos emergentes. Apesar de o monitoramento ser essencial para um planejamento adequado dos recursos hídricos e estratégias de gestão, uma abordagem metodológica abrangente é, em geral, demorada e cara, o que desencoraja o tomador de decisão a fazê-lo. O objetivo deste estudo foi identificar os parâmetros de qualidade da água mais representativos de uma bacia urbana crítica, incluindo concentrações de compostos emergentes. Os resultados destacaram a DBO e as séries de nitrogênio e fósforo como parâmetros químicos; e, condutividade, turbidez e sólidos dissolvidos totais como físicos. Concluiu-se que esses parâmetros representam os usos do solo e da água da bacia e podem contribuir para a definição de um monitoramento mais efetivo, subsidiando a tomada de decisão, o planejamento e a gestão. medilas medi las apropriadamente distintas emergentes gestão geral cara fazêlo. fazêlo fazê lo. lo fazê-lo crítica químicos condutividade físicos Concluiuse Concluiu efetivo

ABSTRACT Changes in water quality are a consequence of water and land use dynamics. Measuring these relationships is challenging, especially when looking at distinct variables such as traditional organic matter constituents and emerging compounds. Although monitoring is necessary for proper water resource planning and management strategies, a comprehensive methodological approach is in general time-consuming and costly, which discourages decision-makers. The goal of the study was to establish consistent water quality elements based on a statistical analysis to identify the most representative parameters of a critical urban basin. The results highlighted BOD, nitrogen, and phosphorus series as chemical representative parameters; and conductivity, turbidity, and total dissolved solids as the physical parameters. In this context, these parameters represent the most significant uses of the studied area and define a more effective monitoring system that will subsidize decision-making and water resources planning and management. dynamics challenging compounds strategies timeconsuming time consuming costly decisionmakers. decisionmakers decision makers. makers decision-makers basin BOD nitrogen conductivity turbidity context decisionmaking making

Abstract Desmoid tumor is an uncommon form of fibromatosis of extremely rare musculoaponeurotic origin. Most are located in the abdomen, and its location in the chest wall is very rare. Due to its inability to produce metastases it has been considered benign, however, its rapid and aggressive local growth makes the surgical treatment equally aggressive and complete. We present a case of a thoracic wall desmoid tumor in a 40-year-old patient, one year evolution, with rapid and aggressive growth which deforms the anatomy of the thoracic wall. Excision of the tumor and 5 partially costal arches were performed, with subsequent thoracic reconstruction with prosthetic material and dermo-fat flap. Desmoid tumors, in advanced stages, requires multidisciplinary participation.

Resumen El tumor desmoide es una forma infrecuente de fibromatosis de origen musculoaponeurótico extremadamente raro. En su mayoría, se localiza en abdomen y es muy raro en pared torácica. Dada su incapacidad para metastatizar se ha considerado benigno, sin embargo, su crecimiento local rápido y agresivo hace que el tratamiento quirúrgico sea igualmente agresivo y completo. Presentamos un caso de tumor desmoide de pared torácica intervenido en paciente de 40 años, de 1 año de evolución y crecimiento rápido y agresivo con deformidad de pared torácica. Realizamos exéresis del tumor y de 5 arcos costales parcialmente, más reconstrucción con material protésico y colgajo dermograso. La resolución quirúrgica de este tipo de tumores en etapas avanzadas requiere participación interdisciplinaria.

Abstract Background: Guillain-Barré syndrome (GBS) patients who present very early (< 4 days) to health-care services generally have severe clinical forms due to rapid progression. Little information exists on the clinical and short-term functional prognosis in patients with very-early GBS (VEGBS). Methods: A prospective cohort study was conducted. We performed Kaplan–Meier survival analysis to observe the recovery of independent gait between groups, where a log-rank value < 0.05 was considered significant. All statistical analyzes were performed using SPSS version 22. Results: Ninety-nine patients were included in the study. Mean age was 46.7 ± 16.7 years, 61.9% were male. Regarding nerve conduction studies, the most frequent electrophysiological variants in the very early-onset GBS group were axonal (38.7%). Twenty-three percent were classified as equivocal. In the GBS group > 4 days, the most frequent electrophysiological variant was acute inflammatory demyelinating polyneuropathy (52.1%). Conclusion: Patients with VEGBS do not present different clinical characteristics, nor severity or poor short-term outcome in our population compared to different GBS presentations. Only mechanical ventilation requirement and distal compound muscle action potential reduction are associated with poor outcome in this subset of patients.

Resumen Antecedentes: Los pacientes con síndrome de Guillain Barré (SGB) que se presentan muy temprano (menos de 4 días) a los servicios de salud generalmente tienen formas clínicas graves debido a la rápida progresión. Existe poca información sobre el pronóstico clínico y funcional a corto plazo en pacientes con presentación ultra temprana. Métodos: Se realizó un estudio de cohorte prospectivo. Realizamos un análisis de supervivencia de Kaplan-Meier para observar la recuperación de la marcha independiente entre los grupos, donde un valor de log-rank < 0.05 se consideró significativo. Todos los análisis estadísticos se realizaron con SPSS versión 22. Resultados: Se incluyeron 99 pacientes. La media de edad fue de 46.7 ± 16.7 años, el 61.9% eran varones. En cuanto a los estudios de conducción nerviosa, las variantes electrofisiológicas más frecuentes en el grupo de SGB de inicio muy precoz fueron axonales (38.7%). El 23% se clasificó como equívoco. En el grupo de SGB > 4 días, la variante electrofisiológica más frecuente fue AIDP (52.1%). Conclusión: Los pacientes con presentación ultratemprana no presentan características clínicas diferentes, ni gravedad ni mala evolución a corto plazo en nuestra población en comparación con diferentes presentaciones de SGB. Solo el requerimiento de ventilación mecánica y la reducción de la CMAP distal se asocian con un resultado deficiente en este subconjunto de pacientes.

Abstract: The wide use of cameras enables the availability of a large amount of image frames that can be used for people counting or to monitor crowds or single individuals for security purposes. These applications require both, object detection and tracking. This task has shown to be challenging due to problems such as occlusion, deformation, motion blur, and scale variation. One alternative to perform tracking is based on the comparison of features extracted for the individual objects from the image. For this purpose, it is necessary to identify the object of interest, a human image, from the rest of the scene. This paper introduces a method to perform the separation of human bodies from images with changing backgrounds. The method is based on image segmentation, the analysis of the possible pose, and a final refinement step based on probabilistic relaxation. It is the first work we are aware that probabilistic fields computed from human pose figures are combined with an improvement step of relaxation for pedestrian segmentation. The proposed method is evaluated using different image series and the results show that it can work efficiently, but it is dependent on some parameters to be set according to the image contrast and scale. Tests show accuracies above 71%. The method performs well in other datasets, where it achieves results comparable to state-of-the-art approaches.

Abstract: This study investigates the potential use of close-range and low-cost terrestrial RGB imaging sensor for fruit detection in a high-density apple orchard of Fuji Suprema apple fruits (Malus domestica Borkh). The study area is a typical orchard located in a small holder farm in Santa Catarina’s Southern plateau (Brazil). Small holder farms in that state are responsible for more than 50% of Brazil’s apple fruit production. Traditional digital image processing approaches such as RGB color space conversion (e.g., rgb, HSV, CIE L*a*b*, OHTA[I 1 , I 2 , I 3 ]) were applied over several terrestrial RGB images to highlight information presented in the original dataset. Band combinations (e.g., rgb-r, HSV-h, Lab-a, I” 2 , I” 3 ) were also generated as additional parameters (C1, C2 and C3) for the fruit detection. After, optimal image binarization and segmentation, parameters were chosen to detect the fruits efficiently and the results were compared to both visual and in-situ fruit counting. Results show that some bands and combinations allowed hits above 75%, of which the following variables stood out as good predictors: rgb-r, Lab-a, I” 2 , I” 3 , and the combinations C2 and C3. The best band combination resulted from the use of Lab-a band and have identical results of commission, omission, and accuracy, being 5%, 25% and 75%, respectively. Fruit detection rate for Lab-a showed a 0.73 coefficient of determination (R2), and fruit recognition accuracy rate showed 0.96 R2. The proposed approach provides results with great applicability for small holder farms and may support local harvest prediction.

Abstract: Object detection in high resolution images is a new challenge that the remote sensing community is facing thanks to introduction of unmanned aerial vehicles and monitoring cameras. One of the interests is to detect and trace persons in the images. Different from general objects, pedestrians can have different poses and are undergoing constant morphological changes while moving, this task needs an intelligent solution. Fine-tuning has woken up great interest among researchers due to its relevance for retraining convolutional networks for many and interesting applications. For object classification, detection, and segmentation fine-tuned models have shown state-of-the-art performance. In the present work, we evaluate the performance of fine-tuned models with a variation of training data by comparing Faster Region-based Convolutional Neural Network (Faster R-CNN) Inception v2, Single Shot MultiBox Detector (SSD) Inception v2, and SSD Mobilenet v2. To achieve the goal, the effect of varying training data on performance metrics such as accuracy, precision, F1-score, and recall are taken into account. After testing the detectors, it was identified that the precision and recall are more sensitive on the variation of the amount of training data. Under five variation of the amount of training data, we observe that the proportion of 60%-80% consistently achieve highly comparable performance, whereas in all variation of training data Faster R-CNN Inception v2 outperforms SSD Inception v2 and SSD Mobilenet v2 in evaluated metrics, but the SSD converges relatively quickly during the training phase. Overall, partitioning 80% of total data for fine-tuning trained models produces efficient detectors even with only 700 data samples.

Abstract: Building roof extraction has been studied for more than thirty years and it generates models that provide important information for many applications, especially urban planning. The present work aimed to model roofs only from point clouds using genetic algorithms (GAs) to develop a more automatized and efficient method. For this, firstly, an algorithm for edge detection was developed. Experiments were performed with simulated and real point clouds, obtained by LIDAR. In the experiments with simulated point clouds, three types of point clouds with different complexities were created, and the effects of noise and scan line spacing on the results were evaluated. For the experiments with real point clouds, five roofs were chosen as examples, each with a different characteristic. GAs were used to select, among the points identified during edge detection, the so-called ‘significant points’, those which are essential to the accurate reconstruction of the roof model. These points were then used to generate the models, which were assessed qualitatively and quantitatively. Such evaluations showed that the use of GAs proved to be efficient for the modeling of roofs, as the model geometry was satisfactory, the error was within an acceptable range, and the computational effort was clearly reduced.

Abstract: The use of OBIA for high spatial resolution image classification can be divided in two main steps, the first being segmentation and the second regarding the labeling of the objects in accordance with a particular set of features and a classifier. Decision trees are often used to represent human knowledge in the latter. The issue falls in how to select a smaller amount of features from a feature space with spatial, spectral and textural variables to describe the classes of interest, which engenders the matter of choosing the best or more convenient feature selection (FS) method. In this work, an approach for FS within a decision tree was introduced using a single perceptron and the Backpropagation algorithm. Three alternatives were compared: single, double and multiple inputs, using a sequential backward search (SBS). Test regions were used to evaluate the efficiency of the proposed methods. Results showed that it is possible to use a single perceptron in each node, with an overall accuracy (OA) between 77.6% and 77.9%. Only SBS reached an OA larger than 88%. Thus, the quality of the proposed solution depends on the number of input features.

Resumo: Avanços na tecnologia LIDAR proporcionaram o desenvolvimento de sistemas que digitalizam completamente o sinal de retorno do pulso LASER emitido. Quando o pulso atinge parcialmente diferentes alvos, múltiplos retornos podem ser determinados dependendo da metodologia utilizada para a detecção dos picos contidos no sinal de retorno. Neste estudo apresentamos os resultados obtidos através da decomposição Gaussiana da waveform do sinal LASER em metodologia que emprega algoritmos genéticos para a determinação dos retornos. Inicialmente a quantidade de picos contidos na waveform é determinada com o método de simples detecção de picos através de algoritmo com ponto de máximo local. Quando mais de um pico é detectado, o processamento é realizado com algoritmos genéticos e são estimados os valores dos parâmetros de amplitude, tempo e desvio padrão. Com esta metodologia foi possível aumentar a quantidade de pontos em aproximadamente 17 % em comparação com a nuvem de pontos gerada em um software comercial. Os melhores resultados foram obtidos em áreas com vegetação alta, comprovando que a metodologia pode ser aplicada para a geração de nuvens de pontos mais densas em áreas de florestas.

Abstract: Recent developments in LIDAR technology lead to the availability of the waveform systems, which capture and digitize the whole return of the emitted LASER pulse. As many objects may cause multiple returns in the same echo, one task is to detect and separate different echoes within the same digitized measurement. In this paper the results of a study aimed at LASER signal waveform decomposition using genetic algorithms are introduced. The proposed method is based on the Gaussian decomposition approach and analyzes each digitized return to compute one or more points. Initially, the number of peaks contained in the waveform is determined by a simple peak detection method, with a local maximum point algorithm. When more than one peak is detected, genetic algorithms are applied to estimate the amplitude, time and standard deviation of each peak within the digitized signal. With this methodology it was possible to increase the number of points by approximately 17 % compared to the point cloud obtained using commercial software. The best results were obtained in areas with high vegetation, and thus the methodology can be applied to the generation of denser points cloud in forest areas.

Técnicas de Sensoriamento Remoto tem ganhado especial interesse, uma vez que podem ser utilizadas para o monitoramento de sistemas e fenômenos em escala local ou global, de maneira contínua temporal e espacialmente. Redes Neurais Artificias estão entre os métodos que são capazes de trabalhar com grande quantidade de dados, com diversas características e sofrer pouca influência de ruídos. Desta forma, fez-se o uso de Redes Neurais Artificiais com o propósito de classificar dados de sensoriamento remoto. Utilizou-se de dados de alta resolução espacial, como imagens espectrais de aerolevantamento e dados altimétricos Laser Scanner, para a classificação do alvo “árvores”. Com isso, gerou-se RNA especialistas na detecção destes alvos. Os dados utilizados são de uma área densamente urbanizada, onde existe grande variabilidade de cotas e características espectrais. Os resultados mostraram que a classificação utilizando dados espectrais e altimétricos resultaram em melhores classificações, do que a utilização apenas de informações espectrais. Testou-se também a influência do tamanho das amostras de treinamento das Redes Neurais Artificiais, gerando assim uma “curva de aprendizado” das RNA. Percebeu-se que conforme se aumenta o tamanho das amostras de treinamento, existe uma tendência em aumentar a acurácia na classificação dos dados. Os acertos globais foram superiores a 87,5% quando utilizando apenas informação espectral e 97,5% quando utilizando dados espectrais e altimétricos

Remote Sensing techniques has gained special interest, since it can be used for monitoring systems and phenomena in local or global scale, in a temporally and spatially continuous way. Artificial Neural Networks are able to work with large amounts of data, with different characteristics. ANN was used in this work as the purpose of classifying remote sensing data. It was used multi sources and high-resolution spatial data, such as spectral images and Laser Scanner elevation data to classify the class “tree”. So, all the ANN created were specialist in tree class classification. In addition, the data used is from a densely urbanized area where there is a large variability of elevations and spectral characteristics. The results showed that the classification using both spectral and altimetry data resulted in better classifications than the use of only spectral information. It was also tested the influence of the size of samples for training the ANN, creating a learning curve for the ANN. It was noticed that with increasing the size of training samples there is a tendency to increase the accuracy in the classification. The global hits were above 87.5% when using only spectral data, and 97.5% when using spectral and altimetry data

O acervo histórico de imagens Landsat permite realizar estudos multitemporais usando uma longa série de imagens. Para isto, as imagens devem ser registradas, ou seja, sua geometria deve ser compatibilizada. Esta tarefa é geralmente executada usando pontos de apoio visíveis nas duas imagens para estimar o modelo de transformação. A escolha e detecção visual de pontos homólogos é uma tarefa árdua e trabalhosa. Neste trabalho é descrito um método automático para a identificação de pontos homólogos usando a correlação entre pares de imagens digitais. Como a correlação pode identificar falsos positivos, ou seja, pontos com alta correlação, mas fora da posição correta, este resultado é refinado analisando a relação espacial entre os pontos disponíveis. Nos experimentos foi utilizada uma área com grande diversidade na paisagem, com agricultura intensiva, malha urbana densa e estradas em grande quantidade, além de um rio, permitindo assim se obter pontos de controle com alvos diversificados. Os resultados obtidos mostram que o processo pode ser automatizado, mas que o sucesso do processo depende da qualidade, em termos de contraste, das regiões usadas na correlação

The historical archive of Landsat imagery allows performing multitemporal studies using a long series of images. For this purpose, images must be registered, that is its geometry must be made compatible. This task generally is performed using visible control points in the two images to estimate the transform parameters. The choice and visual detection of homologous points are arduous and laborious tasks. In this work an automatic method for the identification of homologous points is described, which uses the correlation and spatial analysis. As the digital correlation can identify false positives, that is, points with high correlation but that are far from the right position. This result needs a fine analyzing to select the right points, which is here performed analyzing the spatial relationship between the available points. In the experiments, an area with great diversity in the landscape was used, with intensive agriculture, dense urban occupation, roads and a river, allowing choosing and studying different control points. The results show that the process can be automated, but that the success of the process depends on the quality, in terms of contrast, of the used regions in the correlation

As câmeras de distância são capazes de medir a distância entre o sensor e a superfície dos objetos para cada pixel da imagem, em um único instante, produzindo uma imagem de distâncias. Além das distâncias, também obtém a intensidade e a amplitude para cada pixel, formando as imagens de intensidade e amplitude, respectivamente. Como toda a medida, aquelas obtidas pelas câmeras de distância possuem erros, e um dos fatores que geram erros nas medidas é a iluminação ambiente. Neste estudo foi analisada a influência que a iluminação ambiente exerce sobre os valores calculados de distância, amplitude e intensidade. Para isso, realizou-se experimentos em uma sala escura, onde foram obtidas imagens de distância, intensidade e amplitude de uma parede branca, em que incidia diversos níveis de iluminação gerada por duas lâmpadas cujos feixes convergiam ao centro da parede. Verificou-se que: a iluminação fornecida nos experimentos aumentou os valores de distância; a relação entre os erros de distância e a variação da amplitude é diretamente proporcional e linear; e que a relação entre os erros de distância e a variação da intensidade também se mostrou linear para valores de intensidade menores que 26600, para a frequência de modulação de 21 MHz. As correlações lineares atingiram valores acima de 0,9. A imagem de intensidade se mostrou mais sensível à incidência de iluminação ambiente do que a imagem de amplitude.

Range cameras are able to measure the distance between the sensor and the surface of objects for each image pixel, in a just single instant, producing an image of distance. It also measure the intensity and amplitude for each pixel, in order to produce images of intensity and amplitude, respectively. Like any measure, those obtained by the distance of cameras have errors, and one of the factors that produces errors in the measurements is the external lighting at the environment. In this study it was analyzed the influence of the external lighting on the calculated values of distance, amplitude and intensity. In a dark room it was made up some experiments, where it was obtained images of distance images, intensity and amplitude of a white wall, where various lighting levels generated by two lamps whose beams converged on the center of the wall. Some conclusions are: the external illumination increased distance values; the ratio of the distance error and the amplitude variation is directly proportional and linear; and the relationship between the distance error and the variation of intensity it was also linear for values lower intensities than 26600, at 21 MHz of modulation frequency. It was achieved linear correlation above 0.9. The image of intensity is more sensitive to the effect of external lighting than the amplitude image

As câmaras de distância são capazes de medir a distância entre o sensor e a superfície dos objetos para cada pixel da imagem. Comparando com os equipamentos de varredura a laser possuem a vantagem de obter a distância de vários pontos em um único instante, sem equipamento de varredura. As medidas obtidas pela câmara possuem erros sistemáticos que devem ser minimizados. Alguns fatores como o tempo de integração, a distância a ser medida, bem como a iluminação da cena, influenciam na obtenção da medida. Neste estudo foi analisada a influência da variação do tempo de integração e da distância câmara-alvo na exatidão do cálculo da distância, procurando modelar os erros sistemáticos das medições feitas com uma câmara PMD Camcube 3.0. A modelagem foi feita por meio da Transformada Discreta de Fourier e permitiu diminuir o erro médio quadrático (RMSE) de 15,01 cm para 5,05 cm, para das observações feitas com tempo de integração de 4000 (s. Também foi verificado que a amplitude do erro é diretamente proporcional ao tempo de integração utilizado.

Range cameras are devices to measure the distance between the sensor and the object surface for each pixel within an image. Comparing to laser scanner they have the advantage of measuring range for several points at the same time without scanning systems. The measurements obtained by a range camera may have systematic errors that must be modeled to correct the measurements. Some factors as the integration time used, the distance to be measured, and the illumination of the scene, may affect the range measurement. In this paper we examine the influence of the integration time and the distance between the camera and the targets on the measured distance, proposing a model for the cyclic error of a Camcube PMD 3.0 camera using the Discrete Fourier Transform. The model allowed to reduce the root mean square error (RMSE) from 15.01 cm to 5.05 cm in observations made with the integration time of 4000 (s. It was also verified that the error amplitude is proportional to the integration time.