摘要
基于深度学习在图像分类领域的优异性能,提供一个包含多个苹果品种原始数据集,通过使用ResNet50、InceptionV3以及DenseNet121预训练模型进行苹果品种识别分类,对比3种模型的准确性、泛化能力和稳定性,以期为推进苹果产业标准化、规模化、智能化生产提供理论依据和技术支持。以甘肃省平凉市静宁县果树果品研究所苹果良种苗木繁育基地作为实验基地,采集7个品种共7763幅苹果果实图像自建数据集,进而利用神经网络建立分类识别模型。实验结果验证了卷积神经网络在苹果品种图像识别任务上的优越性能,测试集上表现较优的ResNet50识别准确率达到98.44%,验证了本文方法基于苹果果实进行品种识别与鉴定的有效性。
Based on the excellent performance of deep learning in the field of image classification,we provide a raw dataset containing several apple varieties,and compare the accuracy,generalization ability and stability of the three models by using ResNet50,InceptionV3 and DenseNet121 pre-trained models for apple variety recognition and classification,in order to promote the standardization,scale and intelligence of apple industry.This study aims to provide theoretical basis and technical support for promoting standardized,large-scale and intelligent production in apple industry.A total of 7763 apple fruit images of 7 varieties were collected from the apple seedling breeding base of Jingning County fruit tree and fruit research institute in Pingliang City,Gansu Province as the experimental base,and then a classification recognition model was established using neural networks.The experimental results verified the superior performance of convolutional neural networks for apple variety image recognition task,and the recognition accuracy of ResNet50,which performed better on the test set,reached 98.44%,which verified the effectiveness of this paper's method for variety recognition and identification based on apple fruits.
作者
马斌
韩俊英
MA Bin;HAN Junying(College of Information Science and Technology Gansu Agricultural University,Lanzhou Gansu 733070)
出处
《软件》
2022年第6期17-21,共5页
Software
基金
甘肃省自然科学基金资助项目(20JR5RA023)
甘肃农业大学青年导师基金资助项目(GAU-QDFC-2019-04)。
关键词
图像分类
苹果品种
卷积神经网络
预训练
深度学习
image classification
apple variety
convolutional neural network
pretrained
deep learning
