Neural-Based Hierarchical Approach for Detailed Dominant Forest Species Classification by Multispectral Satellite Imagery

Illarionova, Svetlana; Trekin, Alexey; Ignatiev, Vladimir; Oseledets, Ivan

doi:10.1109/jstars.2020.3048372

Cited by 30 publications

(23 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…With regard to working with multispectral images, methods for use in their visualization [36], restoration [37,38], and correction [29] are being actively developed. As for information retrieval from satellite data, specific problem-oriented methods are being investigated, such as cropland boundaries segmentation [39], crops [40] and tree species classification [41,42], texture classification for forests inventory purposes [43], and windthrow detection [44].…”

Section: Related Workmentioning

confidence: 99%

Segmentation of Agricultural Parcels in Satellite Images Based on Historical Vegetation Index Data

Pavlova¹,

Timofeev²,

Bocharov³

et al. 2021

Preprint

View full text Add to dashboard Cite

This paper considered the issue of agricultural fields boundary recognition in satellite images. A novel algorithm based on the aggregated history of vegetation index data obtained via open satellite data, Sentinel-2, was proposed. The proposed algorithm included several basic steps, namely the detection of parcel regions on aggregated index data; the calculation of aggregated edge maps; the segmentation of parcel regions using the edges obtained; the computation of connected components and their contour extraction. In this paper, we showed that the use of aggregated vegetation index data and boundary maps allow for much more accurate agricultural field segmentation compared to the instant vegetation index approach. The quality of segmentation within regions of Russia and the Ukraine was estimated. The dataset that was used and Python implementation of the proposed algorithm were provided.

show abstract

Section: Related Workmentioning

confidence: 99%

Segmentation of Agricultural Parcels in Satellite Images Based on Historical Vegetation Index Data

Pavlova¹,

Timofeev²,

Bocharov³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…The more parameters the model has, the more complex features it can learn [8], but only with enough training samples available [9]. Indeed, there are many CV datasets both in the general domain [10] and in domains such as autonomous vehicles [11], remote sensing [12], medicine [13], precision agriculture [14], environmental study [15], etc. However, datasets cannot cover all the existing tasks for every specific problem, and a data scientist must prepare data for every new problem [16].…”

Section: Introductionmentioning

confidence: 99%

XtremeAugment: Getting More From Your Data Through Combination of Image Collection and Image Augmentation

et al. 2022

Self Cite

View full text Add to dashboard Cite

Deep convolutional neural networks, being very efficient for computer vision tasks using much training data, still struggle with small training datasets. Therefore, we need a training pipeline that handles rare object types and an overall lack of training data to build well-performing models that provide stable predictions. This paper presents a comprehensive framework XtremeAugment that provides an easy, reliable, and scalable way to collect image datasets and efficiently label and augment collected data. The presented framework consists of two augmentation techniques that can be used independently and complement each other when used together. They are, namely, Hardware Dataset Augmentation (HDA) and Object-Based Augmentation (OBA). HDA is a technique that allows collecting more data and spending less time on manual data labeling. OBA significantly increases training data variability and remains the distribution of the augmented images close to the original dataset. We check the performance of the proposed approach and its independent parts on the apple spoil segmentation problem. Our results show a substantial increase in model accuracy, reaching 0.91 F1-score and outperforming the baseline model on up to 0.62 F1-score for a few-shot learning case in the wild data. The highest benefit from the XtremeAugment raises on the cases where we collect images in the controlled indoor environment but have to use the model in the wild.

show abstract

“…The classical approaches in landcover classification tasks often use NIR-based spectral indices such as the Normalized Difference Vegetation Index (NDVI) or the Enhanced Vegetation Index (EVI) to assess the vegetation state [ 2 ]. This spectral band is widely used in many applications, including forestry [ 3 , 4 ], agriculture [ 5 , 6 ], and general landcover classification [ 7 , 8 ]. However, there are still cases when the NIR band is not presented in the available data [ 9 , 10 ].…”

Section: Introductionmentioning

confidence: 99%

Generation of the NIR Spectral Band for Satellite Images with Convolutional Neural Networks

Illarionova

Shadrin

Trekin

et al. 2021

Sensors

Self Cite

View full text Add to dashboard Cite

The near-infrared (NIR) spectral range (from 780 to 2500 nm) of the multispectral remote sensing imagery provides vital information for landcover classification, especially concerning vegetation assessment. Despite the usefulness of NIR, it does not always accomplish common RGB. Modern achievements in image processing via deep neural networks make it possible to generate artificial spectral information, for example, to solve the image colorization problem. In this research, we aim to investigate whether this approach can produce not only visually similar images but also an artificial spectral band that can improve the performance of computer vision algorithms for solving remote sensing tasks. We study the use of a generative adversarial network (GAN) approach in the task of the NIR band generation using only RGB channels of high-resolution satellite imagery. We evaluate the impact of a generated channel on the model performance to solve the forest segmentation task. Our results show an increase in model accuracy when using generated NIR compared to the baseline model, which uses only RGB (0.947 and 0.914 F1-scores, respectively). The presented study shows the advantages of generating the extra band such as the opportunity to reduce the required amount of labeled data.

show abstract

Neural-Based Hierarchical Approach for Detailed Dominant Forest Species Classification by Multispectral Satellite Imagery

Cited by 30 publications

References 34 publications

Segmentation of Agricultural Parcels in Satellite Images Based on Historical Vegetation Index Data

Segmentation of Agricultural Parcels in Satellite Images Based on Historical Vegetation Index Data

XtremeAugment: Getting More From Your Data Through Combination of Image Collection and Image Augmentation

Generation of the NIR Spectral Band for Satellite Images with Convolutional Neural Networks

Contact Info

Product

Resources

About