当前位置: X-MOL 学术Comput. Electron. Agric. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Identification of stored grain pests by modified residual network
Computers and Electronics in Agriculture ( IF 8.3 ) Pub Date : 2021-02-05 , DOI: 10.1016/j.compag.2021.105983
Yingying Zhang , Weibo Zhong , Hui Pan

Stored grain pests severely affect both grain production and quality. It is essential to adopt technology to identify its kinds as quickly as possible. The objective of this work was to use deep learning to identify six kinds of common stored grain pests. To reduce the cost of manual identification, a method based on computer vision technology, namely modified dilated residual network (MDRN) was proposed. In the proposed method, diliated convolution was introduced into the residual network to further improve the convolution vision of the model without additional parameters. A total of 8,111 RGB (Red, Green, and Blue) images were acquired and then augmented to 31,481 images. To ensure the objective and fair results of the experiment, the model was verified by K-Fold cross validation. After 5-Fold cross validation, the average value of ACC was 96.72%, TPR was 90.17%, TNR was 98.03%, F1 score was 90.17% and AUC was 97% on the original datasets. Experimental results also demonstrate that compared with other recognition methods for the same datasets, the proposed method has the highest accuracy.



中文翻译:

利用改进的残差网络识别储粮害虫

储存的谷物害虫严重影响谷物的产量和质量。采用技术以尽快识别其种类至关重要。这项工作的目的是利用深度学习来识别六种常见的储存谷物害虫。为了降低人工识别的成本,提出了一种基于计算机视觉技术的方法,即改进的扩展残差网络(MDRN)。在提出的方法中,将残差卷积引入残差网络以进一步改善模型的卷积视觉,而无需其他参数。总共获取了8,111张RGB(红色,绿色和蓝色)图像,然后将其扩展为31,481张图像。为了确保实验的客观公正的结果,通过K-Fold交叉验证对模型进行了验证。经过五折交叉验证后,ACC的平均值为96。在原始数据集上,TPR为72%,TPR为90.17%,TNR为98.03%,F1得分为90.17%,AUC为97%。实验结果还表明,与相同数据集的其他识别方法相比,该方法具有最高的准确性。

更新日期:2021-02-05
down
wechat
bug