针对小目标检测提出了一种基于特征金字塔网络改进的算法。通过引入预测优化模块,并结合感兴趣区域的上下文信息,使得特征信息具有更强的稳健性,同时通过内部级联的多阈值预测网络进行预测,最终实现多尺度多阶段的预测,在保证网络参数基本不变的前提下准确率得到提升。实验结果表明,经标准数据集VOC07+12训练后,所提算法在VOC2007测试中的准确率达到80.9%,具有很好的检测性能。

An improved algorithm based on feature pyramid networks is proposed for small target detection. A prediction optimization module is introduced, which is combined with the context information of the region of interest to make the feature information more robust, multi-threshold prediction networks with internal cascade are predicted, and the multi-scale and multi-stage prediction is realized finally. On the premise that the network parameters are basically unchanged, the accuracy is further improved. The experimental results show that the accuracy of the proposed algorithm reaches 80.9% in the VOC2007 test after the training of the standard data set VOC07+12, which has good detection performance.