Research on Multitarget Recognition and Detection Based on Computer Vision

Table 9

Results analysis.

Results	Analysis
The training results based on (240000) are more accurate	This is because the training amount of 240,000 can make SSD have higher detection accuracy
The training results based on (120000) have some error recognition	The training amount of 120000 is insufficient, so SSD generally uses multiple scale predictions of multiple scale feature maps and aspect ratio in images to determine the range of recognition targets, so as to separate and predict them.