Fast forest fire smoke detection using MVMNet

• DOI: 10.1016/j.knosys.2022.108219
• Publication Date: 2022-01-21T01:58:49Z
• Authors: Yaowen Hu; Jialei Zhan; Guoxiong Zhou; Aibin Chen; Weiwei Cai; Kun Guo; Yahui Hu; Liujun Li
• Keywords: Fire Detection; Feature (linguistics); Pyramid (geometry)

Forest fires are a huge ecological hazard, and smoke is an early characteristic of forest fires. Smoke present only in tiny region images that captured the stages occurrence or when far from camera. Furthermore, dispersal uneven, background environment complicated changing, thereby leading to inconspicuous pixel-based features complicate detection. In this paper, we propose detection method called multioriented based on value conversion-attention mechanism module Mixed-NMS (MVMNet). First, proposed. contrast traditional techniques, includes angle parameter data loading process calculates target's rotation using classification prediction method, which has reference significance for determining direction fire source. Then, address issue inconsistent image input size while preserving more feature information, Softpool-spatial pyramid pooling (Soft-SPP) Next, construct (VAM) joint weighting strategy horizontal vertical directions, can specifically extract colour texture smoke. Ultimately, DIoU-NMS Skew-NMS hybrid nonmaximum suppression methods employed issues false missed Experiments conducted homemade dataset, results demonstrate compared our model's mAP reaches 78.92%, 50 88.05%, FPS 122.