Target detection method based on Transform global and local attention interaction

• Main Authors: CHEN YANG, CHEN SIHAN, WANG KUNFENG, ZHANG SHUQIN
• Format: Patent
• Language: Chinese; English
• Published: 12.07.2022
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; HANDLING RECORD CARRIERS; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention belongs to the field of computer vision, particularly relates to a target detection method based on Transform global and local attention interaction, and aims to solve the problems of low accuracy and precision of a target detection result caused by high calculation cost, high complexity and insufficient global and local interaction of a Transform model. The method comprises the following steps: preprocessing a to-be-processed two-dimensional image; performing window division by taking the image token as a unit; performing local multi-head attention calculation based on a window; performing down-sampling on the local windows, splicing the local windows into a new global window, and performing global multi-head attention calculation; global and local interaction is carried out, so that global information is supplemented to local information; image tokens are merged to obtain multi-scale features, and the multi-scale features are sent to a target detection module to obtain the category and positio