No abstract

One feature of most horticultural crop plants that is biologically relevant to their yield and productivity is total leaf area. However, direct methods of estimation of the leaf area cause damage to the plants, whereas indirect methods such as based on light measurement, demand accuracy in the setup of the measurement procedure, which is specific to each crop. Coffee is one of the most important perennial plants related to worldwide trade, and this demands some ability to estimate the productivity of the crop, as well as all the perennial plants involved in production of agricultural products. This study aims to build a model based on indirect measures to estimate the leaf area in coffee plants using image analysis. Two models were evaluated, one based on the height and width of the canopies, and other based on the area of the digital image of a tree. The results of the models have been compared with the real area of the leaves using the destructive approach with measurement of area of all the leaves using a digital scanner. Comparisons between the models and the real values indicated values of adjusted R 2 of about 0.82 with a model using the height and the width values, and about 0.91 in the second model which used the area projection. The robustness of the model using the height and the width values were tested using data presented in the literature to other cultivars and achieved R 2 = 0.54 with an outlier point and 0.91 without it.Key words: non-destructive method, coffee tree, model Estimativa da área foliar total de culturas perenes por meio de análise de imagens RESUMO A área foliar é um atributo biológico relevante para a produtividade de culturas comerciais. Os métodos diretos de estimação da área foliar causam dano às plantas, enquanto os indiretos, como aqueles baseados na medição da quantidade de luz no interior da planta, exigem ajustes e protocolos de medição específicos para cada tipo de cultura. O cafeeiro é uma das mais importantes plantas perenes relacionadas ao comércio de produtos agrícolas em escala mundial, o que demanda habilidade de estimar sua produtividade, tal como ocorre para as outras culturas perenes. Este trabalho visa construir um modelo que contenha um método indireto de estimativa de área foliar em cafeeiros por meio da análise de imagens. Dois modelos foram analisados, sendo que em um foram usadas a altura e a largura dos dosséis e, no outro, se baseou na área projetada do dossel. Os resultados foram comparados com o método direto, através do qual se retiraram todas as folhas dos cafeeiros o que permitiu observar valores de R 2 ajustado de 0,82 para o modelo em que se usaram a altura e a largura dos dosséis, e de 0,91 para o modelo da área projetada. A robustez do método da altura e largura foi testada usando-se dados de literatura relativos a outra cultivar oferecendo valores de R 2 de 0,54, considerando-se um ponto fora da curva, e de 0,91 sem se considerar este ponto.

Object recognition in 3D point clouds is a challenging task, mainly when time is an important factor to deal with, such as in industrial applications. Local descriptors are an amenable choice whenever the 6 DoF pose of recognized objects should also be estimated. However, the pipeline for this kind of descriptors is highly time-consuming. In this work, we propose an update to the traditional pipeline, by adding a preliminary filtering stage referred to as saliency boost. We perform tests on a standard object recognition benchmark by considering four keypoint detectors and four local descriptors, in order to compare time and recognition performance between the traditional pipeline and the boosted one. Results on time show that the boosted pipeline could turn out up to 5 times faster, with the recognition rate improving in most of the cases and exhibiting only a slight decrease in the others. These results suggest that the boosted pipeline can speed-up processing time substantially with limited impacts or even benefits in recognition accuracy.

No abstract

Correspondences between 3D keypoints generated by matching local descriptors are a key step in 3D computer vision and graphic applications. Learned descriptors are rapidly evolving and outperforming the classical handcrafted approaches in the field. Yet, to learn effective representations they require supervision through labeled data, which are cumbersome and time-consuming to obtain. Unsupervised alternatives exist, but they lag in performance. Moreover, invariance to viewpoint changes is attained either by relying on data augmentation, which is prone to degrading upon generalization on unseen datasets, or by learning from handcrafted representations of the input which are already rotation invariant but whose effectiveness at training time may significantly affect the learned descriptor. We show how learning an equivariant 3D local descriptor instead of an invariant one can overcome both issues. LEAD (Local EquivAriant Descriptor) combines Spherical CNNs to learn an equivariant representation together with plane-folding decoders to learn without supervision. Through extensive experiments on standard surface registration datasets, we show how our proposal outperforms existing unsupervised methods by a large margin and achieves competitive results against the supervised approaches, especially in the practically very relevant scenario of transfer learning.