现代电子技术2026，Vol.49Issue(13)：21-24,33,5.DOI:10.16652/j.issn.1004-373X.2026.13.004
深度学习下复杂图像小目标识别算法
Complex image small object recognition algorithm based on deep learning
摘要
Abstract
Small object recognition in complex imagery suffers from limited feature information,severe background interference,and significant color bias,making it difficult to meet the demands of large-scale applications.Therefore,this paper proposes a complex image small object recognition algorithm based on deep learning.In the algorithm,the gray theory and the maximum white balance algorithm are combined to adaptively balance the color of the image and improve the image balance;the global channel attention modulation and spatial attention extraction are applied to the image features,followed by balanced fusion.Fine-grained query-aware sparse attention is then employed to sparsely sample the fused features.The sampled results are fed into a path aggregation network for feature reconstruction,and small object recognition results are ultimately produced by a classifier.The test results show that the proposed algorithm can realize the adaptive correction of complex image color,and all of the values of the background suppression factor are above 10.117 dB;effective small object details can be retained,and the calibrated small objects can be identified successfully.关键词
复杂图像/小目标识别/深度学习/自适应色彩平衡/注意力调制/稀疏采样/路径聚合网络/背景抑制Key words
complex image/small object recognition/deep learning/adaptive color balance/attention modulation/sparse sampling/path aggregation network/background suppression分类
信息技术与安全科学引用本文复制引用
张宁,杜云明,李微娜..深度学习下复杂图像小目标识别算法[J].现代电子技术,2026,49(13):21-24,33,5.基金项目
黑龙江省基本科研业务费基础研究项目(2019-KYYWF-1386) （2019-KYYWF-1386）
