基于人工智能-多元细粒度WSDAN模型探索小麦不完善粒检测新方法

Exploring a new detection method of wheat unsound kernels based on artificial intelligence multivariate fine-grained WSDAN model

针对目前小麦不完善粒检测完全依靠人眼识别的低效性和误差较大的缺陷,构建了基于多元细粒度卷积神经网络的人工智能检测新方法。开展了人工智能检测新方法与传统人眼检测的对比验证,检测结果基本一致,符合国家标准中对不完善粒检测结果的允差要求,新方法模型对完善颗粒的识别准确率高达99.5%、不完善粒的识别准确率达91.7%,总体准确率高达99.1%。研究证明基于多元细粒度卷积神经网络的人工智能检测新方法具有准确、迅速、高效的特点,能够较好地适应小麦不完善粒检测需求,有望今后扩充到现有的标准体系中,逐步实现小麦不完善粒检测的人工智能化。

In view of the low efficiency and large error of wheat unsound kernels detection completely relying on human eye recognition,a new artificial intelligence detection method based on multivariate finegrained convolution neural network was constructed.The comparison and verification between the new artificial intelligence detection method and the traditional human eye detection were carried out.The detection results were basically consistent,which met the tolerance requirements of national standard for the detection of unsound kernels.The accuracy of the new method model was 99.5%for sound kernels 91.7%for unsound kernels.The overall accuracy was up to 99.1%.The research showed that the new artificial intelligence detection method based on multivariate fine-grained convolution neural network was accurate,rapid and efficient,and could better meet the needs of unsound kernels detection of wheat.It was expected to be expanded to the existing standard system in the future,and gradually realize the artificial intelligence of wheat unsound kernels detection.