Tropical Wood Species Recognition: A Dataset of Macroscopic Images

Abstract

Forests are of incalculable value due to the ecosystem services they provide to humanity such as carbon storage, climate regulation and participation in the hydrological cycle. The threat to forests grows as the population increases and the activities that are carried out in it, such as: cattle rearing, illegal trafficking, deforestation and harvesting. Moreover, the environmental authorities do not have sufficient capacity to exercise strict control over wood production due to the vast variety of timber species within the countries, the lack of tools to verify timber species in the supply chain and the limited available and labelled digital data of the forest species. This paper presents a set of digital macroscopic images of eleven tropical forest species, which can be used as support at checkpoints, to carry out studies and research based on macroscopic analysis of cross-sectional images of tree species such as: dendrology, forestry, as well as algorithms of artificial intelligence. Images were acquired in wood warehouses with a digital magnifying glass following a protocol used by the Colombian Ministry of Environment, as well as the USA Forest Services and the International Association of Wood Anatomists. The dataset contains more than 8000 images with resolution of 640 × 480 pixels which includes 3.9 microns per pixel, and an area of (2.5 × 1.9) square millimeters where the anatomical features are exposed. The dataset presents great usability for academics and researchers in the forestry sector, wood anatomists and personnel who work with computational models, without neglecting forest surveillance institutions such as regional autonomous corporations and the Ministry of the Environment.

Dataset:http://dx.doi.org/10.17632/yzzcbyvgmh.3

Dataset License: Creative Commons Attribution 4.0 International.

1. Summary

Forests have incalculable value due to the ecosystem services that they provide to humanity which transcend the tangible elements that can be extracted from them. Among those services are: carbon storage, climate regulation and participation in the hydrological cycle [1]. Globally, the threat to forests grows as the population increases because larger areas are required for agriculture, land conversion for urbanization and timber demand for paper and lumber. Such wood demand results in higher activity of wood extraction and commercialization, both legal and illegal [2,3]. In Colombia, more than half of its surface is covered by natural forests (51.35%) and of this area, 50% is located in zones where forest extraction activities are carried out [4,5]. According to the Ministry of the Environment [6] and ONF Andina [7] in their report “Use and legality of wood in Colombia”, they estimate that 47% of the wood consumed within the country comes from unknown or illegal sources; such a critical aspect generates a negative environmental and socioeconomic impact.

There is a wide variety of native and planted wood species of great interest to the scientific forestry community, and to governments seeking to regulate and protect their exploitation. Timber species differ slightly around the world and show variability in their visible features due to environmental conditions and other factors. In order to carry out studies on dendrology, forestry, as well as for algorithms on artificial intelligence, research is primarily based on macroscopic analysis of cross-sectional images of tree species [8,9], but despite the high biodiversity in the region, there is a limited availability of labeled digital data of forest species, on which studies of algorithms can be carried out to apply recognition and identification [10,11,12]. In addition, in several countries there is a lack of tools to help verify timber species [13,14,15,16] during any step of the supply chain; and the xylotheques that have large taxonomic samples of species are only physical, not digital which makes field identification difficult. The same is also true for universities that carry out different forestry studies and do not have these wood enclosures.

Therefore, this paper provides a set of digital macroscopic images of eleven tropical forest species existing in warehouses and describes the collecting process carried out to set up the digital repository. Images are in RGB format and from timber cross sections with a resolution of 640 × 480 pixels. Image acquisition was performed with a digital magnifying glass which was set at 3.9 microns per pixel, and an area of 2.5 × 1.9 square millimeters where the anatomical features are exposed for macroscopic analysis. The sample preparation process included the cross-section, which is a standard protocol used by the Colombian ministry of environment, as well as internationally practiced by USA forest services and the International Association of Wood Anatomists (IAWA) [17]; moreover, to improve and regularize color contrast, a soft application of water over the cross section was performed. The generated dataset presents great usability for academics in the forestry sector [18], for teaching and research, for wood anatomists and for personnel who work with computational models [19,20], without neglecting the forest surveillance institutions such as regional autonomous corporations and the environmental authorities.

2. Data Description

Out of all the possible existing timber species or those that can be accessed, eleven (11) forest species were selected, which are the most regularly present in the commercialization processes in the region and come from the Amazonian and Pacific regions of Colombia. These species have been rated as being at high risk for illegal timber trade, based on information from the Cauca Regional Autonomous Corporation and the Ministry of the Environment, which have personnel working in the monitoring process of illegal wood trafficking in those areas. Therefore, the dataset is composed of eleven folders, each one named using the scientific names of each forest species (

Table 1. Contents of the database.

Scientific Name/Folder Name	Family	Common Colombian Name/Global Trade Name	Wood Type	Number of Images
Campnosperma panamensis	Anacardiaceae	Sajo/Orey Wood	Hardwood	823
Cedrela odorata	Meliaceae	Cedro costeño/ Cigarbox cedar	Hardwood	1128
Cedrelinga cateniformis	Fabaceae	Achapo/Cedrorana	Hardwood	1189
Cordia alliodora	Boraginaceae	Nogal cafetero/Laurel	Hardwood	929
Dialyanthera gracilipes	Myristicaceae	Cuángare/Virola/White Cedar	Hardwood	1100
Eucalyptus globulus	Myrtaceae	Eucalipto blanco/Blue gum	Hardwood	1105
Handroanthus chrysanthus	Bignoniaceae	Guayacán amarillo/Roble amarillo/Trumpet Tree	Hardwood	1106
Humiriastrum procerum	Humiriaceae	Chanul/Corozo	Hardwood	1001
Fraxinus uhdei	Oleaceae	Urapan/fresno/Shamel ash	Hardwood	1025
Cupresus lusitanica	Cupressaceae	Cipres/Pino Cipres	Softwood	815
Pinus patula	Pinaceae	Pino patula/Ocote	Softwood	571

), and within each folder, there is a ZIP file, named with the Colombian common name, containing macroscopic digital raw RGB images, which are in jpg format. Every picture name has the date of acquisition, and another zip file has also been added with raw images to complement the dataset.

For completeness, the description of the data is composed of three key names, which are useful to recognize the species within the region (Colombia) and around the world: the scientific name, the common Colombian name/and the Global Trade name (Table 1). The description also includes the family, wood type and the number of images in each folder. The specialists mainly use the scientific name and trade names that can be found in timber databases such as InsideWood [21], International Tropical Timber Organization (ITTO) [22] and others; however, the trade name has several variations within the countries.

This dataset shows different types of RGB images with varied textures that expose important features of each species such as fibers, pores, vessels, and parenchyma (Figure 1), which are essential in the identification and diagnosis of timber species. Pictures are from different parts of the log and the database is being populated with pictures of the same tree species coming from different locations, i.e., Amazon rain forest, Pacific rain forest, among others. Moreover, as shown in the examples, the captured sections have random orientations to add variability and make them more usable in producing automatic learning models. In addition, there are two planted species (Figure 1j,k) that do not exhibit the characteristics of native species.

Parameters of the Data Collection

• Image Resolution: 640 × 480 pixels, based on the CMOS sensor size;
• Magnification: 3.9 µm/pixel (microns per pixel);
• Area of interest: 2.5 mm × 1.9 mm;
• Lighting: artificial provided by the capturing device.

The quality of each image was rated on a scale of one to five according to the number of visible features, the focus of the photograph and the cutting technique quality. Some of the defects that are taken as poor quality criteria in the scale ranking are capturing boundaries or borders in the wood, images with a lack of visible features or focusing problems (Figure 2). Finally, images were ordered by date of capture and their names were generated automatically by the acquisition software, as shown below:

format 1: Day MM DD HH-MM-SS or YYMMDD_HHMM.

format 2: IPC_YYYY-MM-DD-HH-MM-SS.

Figure 2. Some defects found in the images. (a) Dialyanthera gracilipes, (b) Handroanthus chrysanthus, (c) Campnosperma panamensis, (d) Fraxinus uhdei. Cutting problems, boundaries, diffuse, lack of features.

Figure 2. Some defects found in the images. (a) Dialyanthera gracilipes, (b) Handroanthus chrysanthus, (c) Campnosperma panamensis, (d) Fraxinus uhdei. Cutting problems, boundaries, diffuse, lack of features.

3. Methods

Image acquisition followed an adaptation of the standard protocol used by the Colombian Ministry of Environment, as well as that internationally practiced by USA forest services and the International Association of Wood Anatomists (IAWA) [17], which was redesigned to the conditions under which local authorities inspect wood in situ.

3.1. Place of Collection

The selection criteria were implemented by the experts of the environmental division of the state public agency, based on species more commercialized or rated as high risk for illegal extraction in the country.

Samples and images were acquired in warehouses, where timber from the Pacific and Colombian Amazon region is aggregated, commercialized and distributed. Samples were taken by an expert from the environmental and forest governance department of the regional autonomous corporation of Cauca, which facilitated access to warehouses. In addition, a set of validation images of the same species was collected in the xylotheque of the University of Cauca to increase variability in the dataset.

3.2. Sample Preparation

For practical purposes, the sample preparation process consisted of cutting the transverse plane of the timber blocks with a small blade and moistening the sample with water to increase contrast and reduce the variation of physical properties such as color due to the moisture content of the wood, which affects multiple physical properties of timber [23]. The cross-section of the wood was selected as the study surface since its cut exposed most of the anatomical characteristics relevant for its discrimination by macroscopic analysis [24].

3.3. Image Adquisition and Labeling

Macroscopic image acquisition followed a capturing protocol that ensures the display of most of the anatomical characteristics of the timber and, hence, their later usability. Here, area, camera and resolution are some of the most important aspects to safeguard image quality [25]. In this sense, macroscopic images were acquired using a digital magnifying glass with artificial lighting and a fixed magnification of 3.9 microns per pixel which covers an area of 2.5 mm × 1.9 mm, a scale suitable for the observation of pores, fibers and parenchyma (Figure 3).

The acquisition was performed using two software applications: MScopes for mobile applications, and AMcap for USB cameras for PC. Image filenames were created automatically by the software. During collection, species were labeled for species and dates in the warehouses. The quality of each image was also rated on a scale of one to five according to the number of visible features, depending on the focus of the photography and the cutting technique. Debugging based on these criteria allowed the construction of a working dataset with macroscopic images of the cross sections. Then, the dataset was separated into folders by species.

4. User Notes

The database contains hundreds of images of both native and planted forest species which allows macroscopic analyses as well as studies of anatomical characteristics of species (Figure 4). It serves for comparative analyses for species variation and identification and as a support for the development of computerized wood recognition systems based on artificial vision and machine learning.

Users can select the set of images of species that are of interest, including various defects present that may or may not help their specific purposes.

Beneficiaries of the data are wood anatomists interested in analyzing variation inter and intraspecies; forest authorities involved in controlling extraction, identification and the trafficking of timber in the laboratory and the field, up to groups, genera and families; and professionals working in artificial intelligence and machine learning, where the data can serve as training sets and for the testing of computational techniques [26].

Data might be used to conduct dendrological research, studies on environmental aspects and climate variability, and their effects on the anatomical structure of trees. The data may become the beginning of a worldwide digital xylotheque (dataset) with hundreds of samples to access easily, where researchers can broaden taxonomic knowledge. The data may also serve as an input to measure the performance of classification algorithms to build specialized forestry software.