针对传统食品图像识别方法提取特征能力差、准确率差、运行效率差和卷积神经网络识别相似食品图像难度大等问题,提出了一种新的食品图像识别模型China Food-CNN,以实现对食物的精准分类;在多分类损失函数Softmax With Loss的基础上,针...针对传统食品图像识别方法提取特征能力差、准确率差、运行效率差和卷积神经网络识别相似食品图像难度大等问题,提出了一种新的食品图像识别模型China Food-CNN,以实现对食物的精准分类;在多分类损失函数Softmax With Loss的基础上,针对食品图像类间相似性大的问题,提出了最大类间距损失函数(MCSWithLoss),以增大相似类之间的距离,实现相似类的区分;针对随机选取样本时的训练集冗余问题,在网络模型训练时使用负样本选择算法.实验结果表明:China Food-CNN模型对食品图像的识别准确率达69.2%,分别比Alex Net、VGG16、Res Net模型提升了17.6%、16.8%和3.6%.展开更多
To retrieve wind field from SAR images, the development for surface wind field retrieval from SAR images based on the improvement of new inversion model is present. Geophysical Model Functions （GMFs） have been widel...To retrieve wind field from SAR images, the development for surface wind field retrieval from SAR images based on the improvement of new inversion model is present. Geophysical Model Functions （GMFs） have been widely applied for wind field retrieval from SAR images. Among them CMOD4 has a good performance under low and moderate wind conditions. Although CMOD5 is developed recently with a more fundamental basis, it has ambiguity of wind speed and a shape gradient of normalized radar cross section under low wind speed condition. This study proposes a method of wind field retrieval from SAR image by com-bining CMOD5 and CMOD4 Five VV-polarisation RADARSAT2 SAR images are implemented for validation and the retrieval re-suits by a combination method （CMOD5 and CMOD4） together with CMOD4 GMF are compared with QuikSCAT wind data. The root-mean-square error （RMSE） of wind speed is 0.75 m s-1 with correlation coefficient 0.84 using the combination method and the RMSE of wind speed is 1.01 m s-1 with correlation coefficient 0.72 using CMOD4 GMF alone for those cases. The proposed method can be applied to SAR image for avoiding the internal defect in CMOD5 under low wind speed condition.展开更多
基金supported by the National Natural Science Foundation of China (Nos.41376010 and 40830959)the Start-up Foundation of Zhejiang Ocean University (No.21105011913)
文摘To retrieve wind field from SAR images, the development for surface wind field retrieval from SAR images based on the improvement of new inversion model is present. Geophysical Model Functions （GMFs） have been widely applied for wind field retrieval from SAR images. Among them CMOD4 has a good performance under low and moderate wind conditions. Although CMOD5 is developed recently with a more fundamental basis, it has ambiguity of wind speed and a shape gradient of normalized radar cross section under low wind speed condition. This study proposes a method of wind field retrieval from SAR image by com-bining CMOD5 and CMOD4 Five VV-polarisation RADARSAT2 SAR images are implemented for validation and the retrieval re-suits by a combination method （CMOD5 and CMOD4） together with CMOD4 GMF are compared with QuikSCAT wind data. The root-mean-square error （RMSE） of wind speed is 0.75 m s-1 with correlation coefficient 0.84 using the combination method and the RMSE of wind speed is 1.01 m s-1 with correlation coefficient 0.72 using CMOD4 GMF alone for those cases. The proposed method can be applied to SAR image for avoiding the internal defect in CMOD5 under low wind speed condition.
