Water channel floater identification method and water channel floater identification model training method

• Main Authors: LI JIANWEN, YU KAI, CHEN XIANLU
• Format: Patent
• Language: Chinese; English
• Published: 05.04.2024
• Subjects: CALCULATING; COMPUTING; COUNTING; PHYSICS

The embodiment of the invention provides a water channel floating object identification model training method, which comprises the steps of processing a to-be-identified water channel image through a feature extraction layer and a feature fusion layer to obtain a fusion feature map of the to-be-identified water channel image, inputting the fusion feature map to a region proposal layer and a main prediction layer to perform prediction processing on a floating object, and inputting the fusion feature map into an auxiliary prediction layer to perform prediction processing on the water channel, and enabling the auxiliary prediction layer to assist training of the main prediction layer by combining a first loss function corresponding to the main prediction layer and a second loss function corresponding to the auxiliary prediction layer, so that the characteristics of the water channel are comprehensively considered. The influence of factors such as width, flow velocity and water surface tension on the motion and f