Automated efficient computation of crown transparency from tree silhouette images

Borianne, Philippe; Subsol, Gérard; Caraglio, Yves

doi:10.1016/j.compag.2016.12.011

“…Image interpretation has to completely rely on texture, viewpoint and lighting variations have to be compensated, and it has to work across a wide range of tree species. Traditional methods that rely on hand-crafted features (Lee et al, 1983;Mizoue, 2002;Dobbertin et al, 2004Dobbertin et al, , 2005Borianne et al, 2017) have to carefully compensate for all disturbing effects with their model design. In contrast, CNNs learn very high-dimensional multi-variate distributions over species and defoliation directly.…”

Section: Methods For Defoliation Estimationmentioning

confidence: 99%

“…Further works of Mizoue (2002) and Dobbertin et al (2004Dobbertin et al ( , 2005 design multi-step image processing workflows that sequentially detect tree regions in the images, extract features, and finally classify often in a semi-supervised way. All methods have in common that they do not apply to the raw image, but need exact delineation of the tree beforehand (Borianne et al, 2017), which is error-prone and often involves user interaction.…”

Section: Related Workmentioning

confidence: 99%

Defoliation estimation of forest trees from ground-level images

Kälin

¹

,

Lang

²

,

Hug

³

et al. 2019

Remote Sensing of Environment

View full text Add to dashboard Cite

In this paper, we propose to estimate tree defoliation from ground-level RGB photos with convolutional neural networks (CNN). Tree defoliation is usually assessed with field campaigns, where experts estimate multiple tree health indicators per sample site. Campaigns span entire countries to come up with a holistic, nation-wide picture of forest health. Surveys are very laborious, expensive, time-consuming and need a large number of experts. We aim at making the monitoring process more efficient by casting tree defoliation estimation as an image interpretation problem. What makes this task challenging is strong variance in lighting, viewpoint, scale, tree species, and defoliation types. Instead of accounting for each factor separately through explicit modelling, we learn a joint distribution directly from a large set of annotated training images following the end-to-end learning paradigm of deep learning. The proposed workflow works as follows: (i) Human experts visit individual trees in forests distributed all over Switzerland, (ii) acquire one photo per tree with an off-theshelf, hand-held RGB camera and (iii) assign a defoliation value. The CNN approach is (iv) trained on a subset of the images with expert defoliation assessments and (v) tested on a hold-out part to check predicted values against ground truth. We evaluate our supervised method on three data sets with different level of difficulty acquired in Swiss forests and achieve an average mean absolute error (avgMAE) of 7.6% for the joint data set after crossvalidation. Comparison to a group of human experts on one of the data sets shows that our CNN approach performs only 0.9 percent points worse. We show that tree defoliation estimation from ground-level RGB images with a CNN works well and achieves performance close to human experts.

show abstract

“…Image interpretation has to completely rely on texture, viewpoint and lighting variations have to be compensated, and it has to work across a wide range of tree species. Traditional methods that rely on hand-crafted features (Lee et al, 1983;Mizoue, 2002;Dobbertin et al, 2004Dobbertin et al, , 2005Borianne et al, 2017) have to carefully compensate for all disturbing effects with their model design. In contrast, CNNs learn very high-dimensional multi-variate distributions over species and defoliation directly.…”

Section: Methods For Defoliation Estimationmentioning

confidence: 99%

“…Further works of Mizoue (2002) and Dobbertin et al (2004Dobbertin et al ( , 2005 design multi-step image processing workflows that sequentially detect tree regions in the images, extract features, and finally classify often in a semi-supervised way. All methods have in common that they do not apply to the raw image, but need exact delineation of the tree beforehand (Borianne et al, 2017), which is error-prone and often involves user interaction.…”

Section: Related Workmentioning

confidence: 99%

Defoliation estimation of forest trees from ground-level images

Kälin

¹

,

Lang

²

,

Hug

³

et al. 2018

Preprint

View full text Add to dashboard Cite

In this paper, we propose to estimate tree defoliation from ground-level RGB photos with convolutional neural networks (CNN). Tree defoliation is usually assessed with field campaigns, where experts estimate multiple tree health indicators per sample site. Campaigns span entire countries to come up with a holistic, nation-wide picture of forest health. Surveys are very laborious, expensive, time-consuming and need a large number of experts. We aim at making the monitoring process more efficient by casting tree defoliation estimation as an image interpretation problem. What makes this task challenging is strong variance in lighting, viewpoint, scale, tree species, and defoliation types. Instead of accounting for each factor separately through explicit modelling, we learn a joint distribution directly from a large set of annotated training images following the end-to-end learning paradigm of deep learning. The proposed workflow works as follows: (i) Human experts visit individual trees in forests distributed all over Switzerland, (ii) acquire one photo per tree with an off-theshelf, hand-held RGB camera and (iii) assign a defoliation value. The CNN approach is (iv) trained on a subset of the images with expert defoliation assessments and (v) tested on a hold-out part to check predicted values against ground truth. We evaluate our supervised method on three data sets with different level of difficulty acquired in Swiss forests and achieve an average mean absolute error (avgMAE) of 7.6% for the joint data set after crossvalidation. Comparison to a group of human experts on one of the data sets shows that our CNN approach performs only 0.9 percent points worse. We show that tree defoliation estimation from ground-level RGB images with a CNN works well and achieves performance close to human experts.

show abstract

“…В статье [9] предлагается метод расчета коэффициента прозрачности кроны по бинарным изображениям деревьев, основанный на автоматическом определении трех показателей: плотности верхней части кроны, плотности макроотверстий и плотности микроотверстий. Вначале извлекается силуэт дерева, затем формируются срезы кроны и строятся профили глубины.…”

Section: анализ существующих методов определения состояния деревьевunclassified

Category Classification of Tree States Using Visual Data Based on Fuzzy Logic

Сергеевич¹,

Николаевна²

2020

ICS

0

View full text Add to dashboard Cite

Introduction: A reliable assessment of tree condition directly affects the planning of economic indicators for the use of forest resources and ecological actions for forest protection. Therefore, the correct evaluation of the sanitary state of forest is very important. At present, the decisions that forest pathologists make about classifying trees or forest areas are based on visual inspection and their subjective knowledge about the tree features. Purpose: Development of a method for classifying the condition of a tree in terms of its crown density degree and other features, based on fuzzy logic with characteristic functions for linguistic variables such as “Crown density”, “Annual branch growth”, “Bark falling off” or “Shrinking branches”. Results: The proposed method classifies the tree condition using pine as an example. The method consists in preliminary image processing, including the removal of background objects, extraction of texture features as extended binary patterns, and application of a specially designed controller based on fuzzy logic. We propose four types of linguistic variables, with their respective terms. For these variables, characteristic functions are specified in tabular form and then approximated by smooth functions. A fuzzy logic controller allows you to obtain an objective assessment of the tree crown condition. Experimental studies confirm the effectiveness of the developed method. Practical relevance: The intelligent system of classifying the tree condition according to visual data can provide a significant support to plantation survey specialists. The proposed method allows you to improve the quality of forest monitoring, minimize the influence of human factor, and organize the forest protection in the best possible way.

show abstract

Automated efficient computation of crown transparency from tree silhouette images

Cited by 8 publications

References 33 publications

Defoliation estimation of forest trees from ground-level images

Defoliation estimation of forest trees from ground-level images

Defoliation estimation of forest trees from ground-level images

Category Classification of Tree States Using Visual Data Based on Fuzzy Logic

Contact Info

Product

Resources

About