基于无人机图像以及不同机器学习和深度学习模型的小麦倒伏率检测

小麦在生长过程中发生倒伏会严重影响其产量,因此实时且准确地对小麦倒伏状况监测有很重要的意义。传统的方法采用手工方式生成数据集,不仅效率低、易出错,而且生成的数据集不准确。针对这一问题,本研究提出了一种基于图像处理的自动数据集生成方法。首先利用无人机在15、46和91 m三个高度采集图像数据;采集完数据后,根据无倒伏、倒伏面积<50%和倒伏面积>50%的标准对每一块地的小麦倒伏情况进行人工评估;采用三种机器学习(支持向量机、随机森林、K近邻)和三种深度学习(ResNet101、GoogLeNet、VGG16)算法对小麦倒伏检测情况进行分类。结果显示,ResNet101的分类结果优于随机森林,并且在91 m高度采集的数据分类精度并不低于在15 m高度采集的数据。本研究证明了针对在91 m高度采集的无人机图像,采用ResNet101对小麦倒伏率检测是一种有效的替代人工检测的方法,其检测精度达到了75%。

Thermal contact conductance estimation and experimental validation in hot stamping process

The variation in temperature of the blank would make major contribution to the mechanical properties of final component in hot stamping of ultra high strength steel(UHSS).It is important to use accurate thermal contact conductance(TCC) to carry out finite element simulation of hot stamping for reliable caculation results.In this paper,a flat compression test was performanced on a servo press.A fast response temperature measurement and data acquisition system was designed to obtain the temperature history of blank and die under different pressures.The thermal contact conductance between blank and die was got using an optimization algorithm.The error between the temperature curves using calculated thermal contact conductance and the curves of measurement temperature was analyzed.Result shows that reliable simulation results of temperature can be got through accurate thermal contact conductance.