Import and export whole link abnormal cargo target identification system based on machine vision

Abstract

The application relates to the technical field of image recognition, and discloses an import and export full-link abnormal cargo target recognition system based on machine vision, which comprises the following steps: acquiring a machine inspection image and obtaining a dual-energy attenuation graph through correction; reversely deducing a container framework coordinate to construct an air connected manifold; then calculating an additional entity attenuation residual in a primary cavity and calculating a local air duct connected blocking amount caused by the residual to air passage continuity; finally, combining the residual and the blocking amount to extract a down-flow cavity invasion parameter, selecting a candidate area with a maximum parameter, and outputting a category probability. The application does not depend on opening a box, effectively improves the detection rate of hidden targets in a cold chain machine inspection image through positioning of abnormal article invasion and blocking performance on air ventilation direction.

Images