Rich feature hierarchies for accurate object detection and semantic segmentation

By A.Y. Virasova, D. Klimov, O.E. Khromov, I. R. Gubaidullin, V. V. Oreshko. 2021

Formulation of the problem. Over the past few years, there has been little progress in object detection techniques. The most efficient are complex computational ensemble methods, which usually combine several low-level image properties with high-level properties. However, every day interest in artificial intelligence is growing, and the idea of using neural networks on board a spacecraft, with the possibility of making decisions and issuing one-time commands, is very promising, since it makes it possible to analyze a large data stream in real time without resorting to ground station, thereby not losing information when transmitting a packet. The purpose of the work is to conduct research on the possibility of effective use of modern models of neural networks, to develop a methodology for their use in the problem of object detection and analysis of the element base for hardware implementation with the possibility of using convolutional neural networks for thermovideotelemetry on board a spacecraft. Results of work. An approach has been formulated that combines two key ideas: 1) application of high-throughput convolutional neural networks for downward processing of image regions in order to localize and segment objects; 2) preliminary training for the auxiliary task, followed by fine tuning of the domain, which gives a significant increase in performance in the case when the training data is insufficient. The analysis of the element base for the possibility of hardware implementation of neural networks on board a spacecraft using electrical radio products of domestic and foreign production is carried out. Practical significance. The efficiency of preliminary network training for an auxiliary task is shown, followed by fine tuning of the subject area. A technique is described that makes it possible to increase the average accuracy of detecting objects in an image by more than 30%. The analysis of the existing element base, the possibility of hardware implementation of neural networks on board the spacecraft using electrical radio products of domestic and foreign production, as well as the criteria for selecting key elements.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

A review on modern defect detection models using DCNNs – Deep convolutional neural networks (2021, methodology): “A comprehensive review for object detection is [4], which thoroughly discusses the subject mentioning the new industry as a whole, starting from non deep-learning object detection methods such as Histogram of Gradients [5] and also discussing in detail the modern approached such as RCNN(region based”
X-DETR: A Versatile Architecture for Instance-wise Vision-Language Tasks (2022, methodology): “One straightforward solution is to use a R-CNN framework [10] with a pretrained object proposal detector, e.”
Few-Shot Adaptive Faster R-CNN (2019, methodology): “Previous detection architectures are grouped into two-or multi-stage models like R-CNN [15], Fast R-CNN [14], Faster R-CNN [37] and Cascaded R-CNN [3], as well as single-stage models like YOLO [35], YOLOv2 [36], SSD [28] and Retinanet [27].”
Enriched Feature Guided Refinement Network for Object Detection (2019, background): “Convolutional neural networks (CNNs) [36, 18, 9, 38, 29, 37] based object detectors [14, 15, 32, 17, 33, 8, 27, 2] have shown outstanding re-sults in recent years.”
IoU Loss for 2D/3D Object Detection (2019, methodology): “R-CNN [13], Fast R-CNN [4], Faster R-CNN [5] and Mask R-CNN [14] are the most representative works of two-stages based methods, while MultiBox [15], YOLO [7], SSD [6], DSSD [16] are the representative works for one-stage based methods.”
Adaptation Across Extreme Variations using Unlabeled Domain Bridges (2019, background): “With advances in supervised deep learning, many vision problems have yielded tremendous improvements [30, 47, 49, 22, 16, 43, 42, 33, 32, 46, 7].”
Recent Advances in Deep Learning for Object Detection (2019, methodology): “tal visual recognition problems: image classification [1] , object de3 tection and instance segmentation [2,3] , and semantic segmenta4”
Single-Shot Bidirectional Pyramid Networks for High-Quality Object Detection (2018, methodology): “BPN320 obtains 354 an accuracy of 80.3%, 75.5% and 66.1% in low, mid and high quality 355 detection scenario respectively, which outperforms many detectors 356 (e.g., SSD320, Faster RCNN, etc.).”
Recent advances and clinical applications of deep learning in medical image analysis (2021, methodology): “Mask RCNN uses a Feature Pyramid Network (FPN) (Setio et al., 2017) as its backbone to extract features at various resolution scales.”
Omni-DETR: Omni-Supervised Object Detection with Transformers (2022, unknown): “Most detection frameworks can be categorized into two groups: two-stage [6,16,17,28,35] v.s. single-stage detectors [30,34,48].”