基于SegNet模型的高原鼠兔的图像分割

针对高原鼠兔图像目标尺寸小、背景复杂、特征不显著、基于活动轮廓的图像分割模型无法有效分割的问题，采用基于卷积神经网络的SegNet语义模型对高原鼠兔图像进行分割：首先将采集的高原鼠兔图像进行预处理，尺度归一化后制作成与Pascal VOC数据集格式一致的数据集；然后将数据集分为训练集与测试集，采用训练集对SegNet模型训练，测试集对模型进行分割测试。对高原鼠兔图像分割的试验结果表明：与基于活动轮廓的Chan_Vese模型相比，基于卷积神经网络的SegNet模型对高原鼠兔图像分割时的交并比、平均像素精度、Dice相似性指数和Jaccard指数分别提高了68.33%、9.35%、30.61%和47.98%，过分割率和欠分割率分别降低了87.20%、16.52%。     

基于深度学习的水稻表型特征提取和穗质量预测研究

水稻产量与水稻穗数和穗质量密切相关,精确预测水稻产量可以加快育种速度。以盆栽水稻为研究对象,首先利用可见光图像结合图像处理技术进行特征提取,获取整株水稻的51个表型特征。结合深度学习,运用Faster R CNN卷积神经网络训练模型对水稻穗数进行检测,同时使用SegNet网络框架训练得到的模型对水稻稻穗进行分割,得到水稻穗部的二值图像,结合图像处理技术提取穗部的33个表型特征数据。提取了颜色、形态、纹理共85个表型参数,对所有85个数据进行归一化处理,将归一化的85个表型数据与稻穗鲜质量、干质量进行逐步线性回归,挑选相关性高的特征数据。分别使用穗数和33个特征穗部、51个特征整株、所有85个特征中相关性高的特征数据构建盆栽水稻稻穗鲜质量、干质量的预测模型,最后根据模型的决定系数R2、平均相对误差（MAPE）和相对误差绝对值的标准差（SAPE）挑选最优预测模型。预测结果表明穗部特征预测效果最好,其中效果最好的模型鲜质量、干质量预测值与真实值的决定系数R2分别达到0.787±0.051和0.840±0.054。     

基于SegNet语义模型的高分辨率遥感影像农村建设用地提取

针对传统分类算法、浅层学习算法不适用于高空间分辨率遥感影像中农村建筑物信息提取的问题,该文以河北省霸州市高空间分辨率遥感影像WorldView-2为数据源,利用182 064幅128×128像素大小的影像切片为训练样本,选取基于深度卷积神经网络的SegNet图像语义分割算法对遥感影像中的农村建筑物进行提取,并与传统分类算法中的最大似然法(maximum likelihood,ML)和ISO聚类、浅层学习算法中的支持向量机(support vector machine,SVM)和随机森林(random forest,RF)以及深层语义分割算法中的金字塔场景解析网络(pyramid scene parsing network,PSPNet)的试验结果作对比分析。研究结果表明:SegNet不仅能够高效利用高空间分辨率遥感影像中农村建筑物的光谱信息而且还能够充分利用其丰富的空间特征信息,最终形成较好的分类模型,该算法在验证样本中的分类总体精度为96.61%,Kappa系数为0.90,建筑物的F1值为0.91,其余5种分类算法的总体精度、Kappa系数、建筑物的F1值都分别在94.68%、0.83、0.87以下。该研究可以为高空间分辨率遥感影像农村建设用地提取研究提供参考。     

Numerical characteristics and spatial distribution of panoramic Street Green View index based on SegNet semantic segmentation in Savannah

Panoramic green view index (PGVI) is an emerging index of urban greenery, which attracts researchers’ attention in recent years. It provides a quantitive method for reflecting real-life feelings about green space in urban areas. The PGVI needs to be calculated from massive datasets, which can be realized by artificial intelligence (AI) techniques. In this work, we used SegNet, an AI semantic segmentation tool, to distinguish urban elements, such as buildings, sky, and people. In total, 6874 panoramic street pictures with an interval of 10 m in the Savannah Historic District were used for the analysis of PGVI and its distribution. Results show that both the PGVI value and its distribution types can reflect the characteristics of regional green space. Good urban greenery can be distributed normally, which also provides a method for greenery classification. The crucial factors influencing PGVI are the trees. Dense low trees with big canopies have a very positive influence. In addition, the grade and width of the road, the parks, and squares along the street also have an impact on PGVI. In Savanah Historic District, the road width of nearly 10 m, and the location near parks and squares, can significantly increase the PGVI of streets.