P‐3.11: An Improved SSD Algorithm
Xianguang Li, Wei Jia, Aoxiang Jia, Jianpu Lin · SID Symposium Digest of Technical Papers · 2025
Compared with Fast R‐CNN [1] , YOLO [2‐4] and other detection algorithms [5] , SSD algorithm [6] has better detection accuracy and real‐time performance detection performance. However, the comprehensive detection performance of SSD algorithm is strong, but it still has some shortcomings in network structure, model training and other aspects. Therefore, aiming at the problems of large network calculation, low multi‐target detection accuracy and unbalanced feature learning in the application of SSD algorithm to multi‐target detection and classification, this paper proposes an Adaptive Feature Fusion (AFF) structure that can be applied to lightweight detection algorithm, which includes Channel Adaptive Weighting (CAW) module and Spatial Feature Aggregation (SFA) module. And based on the optimization of rate strategy learning, the AFSSD‐MV2 (Adaptive FSSD with MobileNetV2) algorithm is proposed.