Multi-scale target detection method introducing attention mechanism

• Main Authors: DAI XIANHUA, KUANG LIJUAN
• Format: Patent
• Language: Chinese; English
• Published: 11.12.2020
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

The invention relates to a multi-scale detection technology introducing an attention mechanism, and relates to the field of image processing, and the method comprises the steps: collecting a to-be-detected image, importing the to-be-detected image into an attention yolo-v3, enabling the attention yolo-v3 to be extended on the basis of the yolo-v3, adding an SENet with a channel attention mechanism, and carrying out the parallel prediction of a target through conventional detection; inputting the picture to be tested into a pre-trained neural network (darknet-53 + FPN) to obtain a feature map of three scales; carrying out clustering to obtain nine prior frames anchor boxs of three scales in total, wherein three candidate boxes bbox can be generated at each point in the featured map grid; calculating IOUs of the prediction frame and the real frame, and allocating an optimal match to the real frame; enabling the candidate bbox to be subjected to classification and border (BBox) regression; and finally, filtering