Small target detection method based on multi-scale information fusion

• Main Authors: JIN QIJIAN, WU YUE, ZHANG HONGTIAN, SHEN ZHIXI, LI JUNBEI
• Format: Patent
• Language: Chinese; English
• Published: 12.09.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention relates to a small target detection method based on multi-scale information fusion, and the method comprises the following steps: obtaining an existing data set, carrying out the preprocessing of the data set, and forming a training set through all images after preprocessing; a small target detection model is constructed, the small target detection model adopts a trunk feature extraction network of YOLOv5 as a basic model, and several modules in the basic model are replaced; data resampling operation is introduced into the data of the input data set, training is carried out, and model parameters are stored after optimal convergence to obtain an optimal small target detection model; and calling the optimal small target detection model, inputting a detected picture for detection, and outputting the detected picture as an identification result. According to the experimental verification result, the method improves the multi-scale feature extraction capability, effectively improves the detection per