KutralNet: A Portable Deep Learning Model for Fire Recognition

Ayala, Angel; Fernandes, Bruno; Cruz, Francisco; Macêdo, David; Oliveira, Adriano L. I.; Zanchettin, Cleber

doi:10.1109/ijcnn48605.2020.9207202

Cited by 21 publications

(7 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The next step is to modify the network architecture of the model to speed up the recognition process without significantly decreasing the accuracy. Ku-ralNet is a lightweight deep learning model that strikes a good balance between parameters and effectiveness [35]. In the KuralNet, the inverse residual block with deep convolution and frequency-doubling convolution can be used for signal processing to reduce the computational cost.…”

Section: Discussionmentioning

confidence: 99%

A Soft-YoloV4 for High-Performance Head Detection and Counting

et al. 2021

View full text Add to dashboard Cite

Blockage of pedestrians will cause inaccurate people counting, and people’s heads are easily blocked by each other in crowded occasions. To reduce missed detections as much as possible and improve the capability of the detection model, this paper proposes a new people counting method, named Soft-YoloV4, by attenuating the score of adjacent detection frames to prevent the occurrence of missed detection. The proposed Soft-YoloV4 improves the accuracy of people counting and reduces the incorrect elimination of the detection frames when heads are blocked by each other. Compared with the state-of-the-art YoloV4, the AP value of the proposed head detection method is increased from 88.52 to 90.54%. The Soft-YoloV4 model has much higher robustness and a lower missed detection rate for head detection, and therefore it dramatically improves the accuracy of people counting.

show abstract

Section: Discussionmentioning

confidence: 99%

A Soft-YoloV4 for High-Performance Head Detection and Counting

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Three publicly available datasets used were designed for a fire or smoke single-label classification task, with fire, smoke, or none classes, named FiSmo 1 [27], FireNet 2 [26], and FireSmoke 3 . All the datasets were previously used in fire and fire and smoke classification tasks as presented in [9], [26], [28], [38], [39]. For this project, 16,140 datasets' images were checked by one person, labeling all the images for a multi-label classification approach.…”

Section: A Datasetsmentioning

confidence: 99%

“…The FireNet dataset contains a test subset with 871 images, used for testing purposes for the fire-only recognition task. The augmentation of FiSmo used in this research adds 485 black images as none class because it has been shown to improve the model's performance for fire recognition [9], [38]. Finally, the combination of the FireNet and FireSmoke datasets were merged into a new one called KutralSmoke with 6, 296 images.…”

Section: A Datasetsmentioning

confidence: 99%

See 1 more Smart Citation

Convolution Optimization in Fire Classification

et al. 2022

Self Cite

View full text Add to dashboard Cite

Early alert fire and smoke detection systems are crucial for daily and security management decision-making. Recent literature approaches are based on Deep Learning (DL) models. Nevertheless, many developments have been focused on surpassing the previous state-of-the-art model's accuracy, regardless of the computational resources needed to execute the model. Efficient models are required for hardware-constrained systems, such as mobile devices, embedded systems, and robotics achieving high performance at low power consumption. For this research, we designed a novel specific-purpose model for fire and smoke recognition using still images and the study of state-of-the-art convolution techniques to improve the trade-off between accuracy and complexity. In this regard, the literature suggests that the inverted residual block, the depthwise and octave convolution techniques, reduces the model's size and computation requirements working well by themselves. In this work, we propose the KutralNext architecture, an efficient model for single-and multi-label fire and smoke recognition tasks. Additionally, a more efficient architecture KutralNext+, demonstrates that those convolution techniques achieve a better trade-off combined, reaching an 84.36% average test accuracy in FireNet, FiSmo, and FiSmoA fire datasets. The KutralSmoke and FiSmo fire and smoke datasets attained an 81.53% average test accuracy. Furthermore, a previous fire and smoke recognition model considered, FireDetection, KutralNext uses 59% fewer parameters, and KutralNext+ requires 97% fewer flops and is 4x faster.

show abstract

“…The following proposal for fire recognition sets a baseline model to develop portable versions focused on reducing the model's complexity in processing the input image. The Kutral-Net 3 [23] model was developed as a suitable option for limited hardware devices and built other efficient versions using the octave convolution and the inverted residual block to test each efficient technique by themselves and combined. Hereof, three portable models were obtained from this baseline using efficient deep learning techniques.…”

Section: B Lightweight Efficient Deep Learning Modelmentioning

confidence: 99%

KutralNext: An Efficient Multi-label Fire and Smoke Image Recognition Model

Ayala¹,

Macêdo²,

Zanchettin³

et al. 2021

Anais Estendidos Da XXXIV Conference on Graphics, Patterns and Images (SIBRAPI Estendido 2021)

Self Cite

View full text Add to dashboard Cite

Early alert fire and smoke detection systems are crucial for management decision making as daily and security operations. One of the new approaches to the problem is the use of images to perform the detection. Fire and smoke recognition from visual scenes is a demanding task due to the high variance of color and texture. In recent years, several fire-recognition approaches based on deep learning methods have been proposed to overcome this problem. Nevertheless, many developments have been focused on surpassing previous state-of-the-art model's accuracy, regardless of the computational resources needed to execute the model. In this work, is studied the trade-off between accuracy and complexity of the inverted residual block and the octave convolution techniques, which reduces the model's size and computation requirements. The literature suggests that those techniques work well by themselves, and in this research was demonstrated that combined, it achieves a better trade-off. We proposed the KutralNext architecture, an efficient model with reduced number of layers and computacional resources for singleand multi-label fire and smoke recognition tasks. Additionally, a more efficient KutralNext+ model improved with novel techniques, achieved an 84.36% average test accuracy in FireNet, FiSmo, and FiSmoA fire datasets. For the KutralSmoke and FiSmo fire and smoke datasets attained an 81.53% average test accuracy. Furthermore, state-of-the-art fire and smoke recognition model considered, FireDetection, KutralNext uses 59% fewer parameters, and KutralNext+ requires 97% fewer flops and is 4x faster.

show abstract

KutralNet: A Portable Deep Learning Model for Fire Recognition

Cited by 21 publications

References 18 publications

A Soft-YoloV4 for High-Performance Head Detection and Counting

A Soft-YoloV4 for High-Performance Head Detection and Counting

Convolution Optimization in Fire Classification

KutralNext: An Efficient Multi-label Fire and Smoke Image Recognition Model

Contact Info

Product

Resources

About