深度学习下复杂图像小目标识别算法OA

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.