Submitted by WALTER MALDONADO JÚNIOR null (walter@rainformatica.com.br) on 2016-03-29T21:27:14Z No. of bitstreams: 1 principal.pdf: 75187969 bytes, checksum: ed5b4271338552ed5f58e72f73d7073d (MD5) / Approved for entry into archive by Ana Paula Grisoto (grisotoana@reitoria.unesp.br) on 2016-03-30T11:37:55Z (GMT) No. of bitstreams: 1 maldonadojunior_w_dr_jabo.pdf: 75187969 bytes, checksum: ed5b4271338552ed5f58e72f73d7073d (MD5) / Made available in DSpace on 2016-03-30T11:37:55Z (GMT). No. of bitstreams: 1 maldonadojunior_w_dr_jabo.pdf: 75187969 bytes, checksum: ed5b4271338552ed5f58e72f73d7073d (MD5) Previous issue date: 2016-02-22 / Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) / A estimativa da produtividade é um fator importante no planejamento de um processo produtivo. No caso dos citros, pode colaborar com o gerenciamento do processo industrial e servir como orientação para os produtores, apresentando papel decisivo no mercado do produto e no manejo de tratos culturais. Vários estudos de técnicas para estimativa da produção da cultura vêm sendo realizados mas ainda apresentando limitações. Devido à correlação entre o número de frutos visíveis na imagem de uma planta e o número real de frutos na mesma já apontada em estudos anteriores, foi desenvolvido um método de amostragem automático e não-destrutivo, por meio da extração das características de frutos verdes em imagens digitais. Utilizou-se uma combinação das técnicas de conversão do modelo de cores, limiarização, equalização do histograma de níveis de cinza, filtragem espacial com os operadores de Laplace e Sobel e suavização gaussiana. Além disso, foi desenvolvido e testado um algoritmo para o reconhecimento e contagem dos frutos nessas imagens, com taxas de detecção de falso-positivos de 3\% em imagens de boa qualidade. É possível se estimar a média do número de frutos visíveis por planta com um erro tolerado de 5\% com até 46 imagens e em aproximadamente 8 minutos, sem nenhuma interação humana. A ausência de flash e a incidência de luz solar direta sobre a planta podem prejudicar consideravelmente o desempenho do algoritmo. / Yield estimation is an important factor in a production process planning. In the case of citrus orchards, can be useful for processing plants management and as guidance for farmers, showing a decisive role in the product market strategies and cultivation practices. Several techniques are being studied for estimating citrus crop yield, but still presenting significant limitations. On the basis of the known correlation between the number of visible fruits in a digital image and the total of fruits present in an orange tree, an automatic and non-destructive method for green fruit feature extraction was developed with a combination of the techniques of color model conversion, thresholding, histogram equalization, spatial filtering with Laplace and Sobel operators and gaussian blur. In addition, we built and tested an algorithm to recognize and count the fruits, with detection rates of false-positives of 3\% for images acquired in good conditions. It is possible to estimate the mean number of visible fruits in the trees within a tolerated error of 5\% with up to 46 images and taking approximately 8 minutes without any human interaction. The absence of flash light or the direct incidence of solar light on the plant can significantly detract the algorithm results. / CNPq: 140600/2013-2

Traditional image processing solutions have been applied in the past to detect and count strawberries. These methods typically involve feature extraction followed by object detection using one or more features. Some object detection problems can be ambiguous as to what features are relevant and the solutions to many problems are only fully realized when the modern approach has been applied and tested, such as deep learning. In this work, we investigate the use of R-CNN for green strawberry detection. The object detection involves finding regions of interest (ROIs) in field images using the selective segmentation algorithm and inputting these regions into a pre-trained deep neural network (DNN) model. The convolutional neural networks VGG, MobileNet and ResNet were implemented to detect subtle differences between green strawberries and various background elements. Downscaling factors, intersection over union (IOU) thresholds and non-maxima suppression (NMS) values can be tweaked to increase recall and reduce false positives while data augmentation and negative hardminging can be used to increase the amount of input data. The state of the art model is sufficient in locating the green strawberries with an overall model accuracy of 74%. The R-CNN model can then be used for crop yield prediction to forecast the actual red strawberry count one week in advance with a 90% accuracy.