Multi-modal marine target detection method based on multi-scale capsule and Bi-FPN

• Main Authors: LIU JIN, LI XINGYE, TANG ZHENYU
• Format: Patent
• Language: Chinese; English
• Published: 13.12.2022
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention discloses a multi-modal marine target detection method based on a multi-scale capsule and Bi-FPN. The method comprises the following steps: S1, carrying out multi-scale feature fusion on RGB-D multi-modal image data of a marine scene through a bidirectional feature pyramid; s2, optimizing the fusion effect of the shallow features through an iterative fusion method; s3, respectively inputting the multi-scale features obtained after feature fusion into a classification network and a regression network based on the multi-scale capsule network in a hierarchical manner; and S4, through a depth separable convolution and dynamic routing algorithm technology, respectively calculating tetrad vectors of the anchor frame category prediction matrix and the target object prediction frame coordinate which are finally required to be obtained. According to the method, multiple deep neural network technologies are fused, the accuracy rate of offshore scene target detection is improved, objects with different siz