Moving vehicle detection in aerial infrared image sequences via fast image registration and improved YOLOv3 network

Weak and Occluded Vehicle Detection in Complex Infrared Environment Based on Improved YOLOv4

“…. Average Precision is a normal evaluation metric for object detection, but in binary classification, to mitigate the influence of imbalance of positive and negative examples, we also combined the F1 score as a comprehensive evaluation metric, as shown in equitation (3)(4)(5). YOLOv4 original model for the testing set has a higher average precision and F1 score compared with the other models, this is due to the superiority of YOLOv4 model structure itself, and the cosine annealing algorithm was used in the process of training enabling the parameter in the process of training to achieve a more optimal solution.…”

“…For infrared object detection, the common infrared targets are infrared pedestrian detection [1,20,21,22], infrared vehicle and aircraft detection [3,4,5,23]. The main problem to be overcome is the lack of infrared data, followed by the problem of unclear infrared image features.…”

“…This network uses feature fusion, which fuses different feature extraction layers to retain small-scale information, and introduces attention mechanism into feature maps to highlight small-scale features. For vehicle detection, Zhang et al [3,4] improved the YOLOv3 model by reducing the numbers of network layers of feature extraction as well as transfer learning, and calculated the appropriate sizes of anchor boxes through k clustering algorithm, so as to detect low-resolution vehicles in infrared aerial images and then track them.…”

“…Firstly, the cascade classifier is applied to identify the eyes and mouth of a human face; then, the physical feature relationship between the eyes and mouth of a person and the face is used to identify and detect the entire masked face. In the context of urban autonomous driving system, Only useful for objects of large scales in simple background [1], [3][4][5], [24] Feature fusion+ attention mechanism Extract features of very small objects(2×2 pixels) Only useful for objects in simple background such as sky and sea [2], [6] Density map Detect the small and high density objects Not useful for single object detection [16] Extra information Has the ability of reasoning and predict the occluded parts Detection on traditional visible datasets [17][18][19] Ours (secondary transfer learning + HNEM)…”

A PCB Electronic Components Detection Network Design Based on Effective Receptive Field Size and Anchor Size Matching

“…Kong et al proposed a model in Darknet-53 that conducts efficient feature extraction via the Dual-Path Network (DPN) module and the fusion transition module during the real-time sonar target detection [ 28 ]. Zhang and Zhu replaced the original Darknet-53 with the Darknet-23 to improve the detection speed when detecting moving vehicles in aerial infrared image sequences [ 29 ]. Li et al employed depthwise separable convolution to design the backbone network to reduce the parameters and the extract crack features effectively for crack inspection in aircraft structures [ 30 ].…”