基于特征融合的玉米品种识别

为了充分利用图像的多级尺度特征，同时发挥深度学习与传统特征各自在提取玉米图像深度特征和底层特征方面的优势，进一步提升玉米品种识别的准确率，提出一种基于特征融合的玉米品种图像识别方法。以苏玉10、京科968和正大619三个玉米品种为研究对象，制作数据集并标记类别标签，分别记为1、2、3。通过VGG16和ResNet50两种预训练网络来获取图像的深度特征，并与人工提取的特征进行融合得到新的玉米图像特征，输入到不同的分类器对玉米图像进行分类。实验结果表明，对特征进行融合相较于单一使用深度特征或传统特征具有更高的识别准确率，在三个玉米品种上分别达到99.58%、98.75%、99.17%，平均准确率为99.17%。

Maize variety recognition based on feature fusion

In order to make full use of the multi-level scale features of the image, give full play to the advantages of depth learning and manual extraction features in extracting the depth features and bottom features of the corn image, and further improve the accuracy of maize variety recognition, a maize variety image classification method based on feature fusion is proposed. Three maize varieties Suyu 10, Jingke 968 and Zhengda 619 were taken as the research objects, and data sets were made and category labels were marked as 1, 2 and 3 respectively. The depth features of the image are acquired through VGG16 and ResNet50 pre-training networks and fused with the manually extracted features to obtain new maize image features, which are input to different classifiers to classify maize images. The experimental results show that compared with the single use of depth features or traditional features, feature fusion has a higher recognition accuracy, reaching 99.58%, 98.75% and 99.17% respectively on three maize varieties, with an average accuracy of 99.17%.