Active Fire Mapping on Brazilian Pantanal Based on Deep Learning and CBERS 04A Imagery

Higa, Leandro Tsuneki; Marcato, José; Rodrigues, Thiago Rangel; Zamboni, Pedro; Silva, Rodrigo; Almeida, Laisa; Liesenberg, Veraldo; Roque, Fábio de Oliveira; Libonati, Renata; Gonçalves, Wesley Nunes; Silva, Jonathan R.

doi:10.3390/rs14030688

Cited by 31 publications

(13 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…However, ultrahigh temporal resolution data such as GOES-16 ABI is thus far poorly explored in this field. The authors of [50] proposed an approach based on object detection methods to map AF in the Brazilian Pantanal biome. For that, the authors used deep learning (a subset of ML-based on neural networks) and CBERS 4A (China Brazil Earth Resources Satellite) imagery.…”

Section: Discussionmentioning

confidence: 99%

Near Real-Time Fire Detection and Monitoring in the MATOPIBA Region, Brazil

Pletsch

Körting

Morita³

et al. 2022

Remote Sensing

View full text Add to dashboard Cite

MATOPIBA is an agricultural frontier, where fires are essential for its biodiversity maintenance. However, the increase in its recurrence and intensity, as well as accidental fires can lead to socioeconomic and environmental losses. Due to this dual relationship with fire, near real-time (NRT) fire management is required throughout the region. In this context, we developed, to the best of our knowledge, the first Machine Learning (ML) algorithm based on the GOES-16 ABI sensor able to detect and monitor Active Fires (AF) in NRT in MATOPIBA. To do so, we analyzed the best combination of three ML algorithms and how long it takes to consider a historical time series able to support accurate AF predictions. We used the most accurate combination for the final model (FM) development. The results show that the FM ensures an overall accuracy rate of approximately 80%. The FM potential is remarkable not only for single detections but also for a consecutive sequence of positive predictions. Roughly, the FM achieves an accuracy rate peak after around 20 h of consecutive AF detections, but there is an important trade-off between the accuracy and the time required to assemble more fire indications, which can be decisive for firefighters in real life.

show abstract

Section: Discussionmentioning

confidence: 99%

Near Real-Time Fire Detection and Monitoring in the MATOPIBA Region, Brazil

Pletsch

Körting

Morita³

et al. 2022

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…The models that performed the best on the validation set were chosen independently to make inferences on the data from the test set. For the inference findings, we employed KBS ( Figure 4B ) to decode them rather than non-maximum suppression (NMS) ( 32 ). First, the category corresponding to the prediction result was divided into the key bone and development grade, and that grade’s confidence level and prediction box were recorded.…”

Section: Methodsmentioning

confidence: 99%

“…In the first part, Both the PP-PicoDet and NanoDet models are anchor-free models, while the YOLOv5 model employed the Kmeans method to obtain anchors such as [[23,24, 27,28, 26,34], [32,33,31,41,37,38], [38,48,54,58,66,69]]. The images were preprocessed before model training, including resizing the images to correspond to the size required by the model (640x640 for YOLOv5, 416x416 for PP-PicoDet, and 416x416 for NanoDet) and normalizing the images to a range of pixel values of (0, 1).…”

Section: Training Modelmentioning

confidence: 99%

A real-time automated bone age assessment system based on the RUS-CHN method

Yang

Dai²,

Qin³

et al. 2023

Front. Endocrinol.

View full text Add to dashboard Cite

BackgroundBone age is the age of skeletal development and is a direct indicator of physical growth and development in children. Most bone age assessment (BAA) systems use direct regression with the entire hand bone map or first segmenting the region of interest (ROI) using the clinical a priori method and then deriving the bone age based on the characteristics of the ROI, which takes more time and requires more computation.Materials and methodsKey bone grades and locations were determined using three real-time target detection models and Key Bone Search (KBS) post-processing using the RUS-CHN approach, and then the age of the bones was predicted using a Lightgbm regression model. Intersection over Union (IOU) was used to evaluate the precision of the key bone locations, while the mean absolute error (MAE), the root mean square error (RMSE), and the root mean squared percentage error (RMSPE) were used to evaluate the discrepancy between predicted and true bone age. The model was finally transformed into an Open Neural Network Exchange (ONNX) model and tested for inference speed on the GPU (RTX 3060).ResultsThe three real-time models achieved good results with an average (IOU) of no less than 0.9 in all key bones. The most accurate outcomes for the inference results utilizing KBS were a MAE of 0.35 years, a RMSE of 0.46 years, and a RMSPE of 0.11. Using the GPU RTX3060 for inference, the critical bone level and position inference time was 26 ms. The bone age inference time was 2 ms.ConclusionsWe developed an automated end-to-end BAA system that is based on real-time target detection, obtaining key bone developmental grade and location in a single pass with the aid of KBS, and using Lightgbm to obtain bone age, capable of outputting results in real-time with good accuracy and stability, and able to be used without hand-shaped segmentation. The BAA system automatically implements the entire process of the RUS-CHN method and outputs information on the location and developmental grade of the 13 key bones of the RUS-CHN method along with the bone age to assist the physician in making judgments, making full use of clinical a priori knowledge.

show abstract

“…Coupled with the ability of the SAR systems of acquiring data regardless of illumination and atmospheric conditions, deep learning methods have shown great potential for extracting patterns of changes in images (Ban et al, 2020;Higa et al, 2022;Zhang et al, 2021), particularly semantic segmentation networks, such as the U-Net, proposed by Ronneberger et al (2015). U-Net is a convolutional neural network architecture that performs semantic segmentation of images.…”

Section: Introductionmentioning

confidence: 99%

Burnt Areas Semantic Segmentation From Sentinel Data Using the U-Net Network Trained With Semi-Automated Annotations

Marra,

Galo,

Giulio Tonolo

et al. 2023

Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

Abstract. The Pantanal biome is one of the most important wetlands on the planet, harboring a rich biodiversity whilst being critical in maintaining hydrological cycles and climate regulation. However, the occurrence of fires in the biome has represented a significant threat to this unique ecosystem and its multiple functions. Understanding the extent, intensity and environmental impacts caused by fires in the Pantanal, is of unique importance for the preservation of the biome's biodiversity. Remote sensing techniques have played an important role in detecting and mapping burnt areas, especially SAR (Synthetic Aperture Radar) orbital systems, that are able to collect data in regions with frequent cloud cover or during extreme fire events. In this context, the objective of this study was to evaluate the potential of the U-Net semantic segmentation network applied to SAR data in the detection of burnt areas in the Brazilian Pantanal. For this, a semi-automatic annotated dataset was generated and considered as ground truth to evaluate the result obtained by the network. Two input datasets were evaluated in the detection of burnt areas, one containing optical and SAR data whereas the other containing only SAR data. The predictions of the two datasets were consistent with the semi-automatically generated annotation, showing similar spatial distribution but presenting a greater number of burnt areas. The model using both optical and SAR data achieved IoU (Intersection of Union) of 0.69 whereas the SAR only model had 0.60. Considering the amount of available data and the complexity of burnt area detection, the predictions achieved were adequate.

show abstract

Active Fire Mapping on Brazilian Pantanal Based on Deep Learning and CBERS 04A Imagery

Cited by 31 publications

References 41 publications

Near Real-Time Fire Detection and Monitoring in the MATOPIBA Region, Brazil

Near Real-Time Fire Detection and Monitoring in the MATOPIBA Region, Brazil

A real-time automated bone age assessment system based on the RUS-CHN method

Burnt Areas Semantic Segmentation From Sentinel Data Using the U-Net Network Trained With Semi-Automated Annotations

Contact Info

Product

Resources

About