基于卷积神经网络的小麦产量预估方法

小麦产量是评估农业生产力的重要指标之一,针对小麦产量人工预估困难,提出将卷积神经网络运用于小麦产量预估,为农业生产力的预估提供参考,指导农业生产管理决策。利用无人机分别在河南省新乡、漯河两地进行图片采集,并以之构建麦穗数据集,分为正样本(麦穗)和负样本(叶子和背景)。针对小麦常规的生理形态和生长环境,设计卷积神经网络识别模型,以图像金字塔构建多尺度滑动窗口,以非极大值抑制(NMS)去除重叠率较高的目标框,实现对单位面积内麦穗的计数,并利用随机采样的方式对大田麦穗进行单位面积图像采样,以采样图像中麦穗数量的平均值作为产量预估基准,进一步实现麦穗产量预估。随机抽取100幅不同小麦图片进行测试,与人工计数结果进行对比,准确率达到97.30%,漏检率为0.34%,误检率为2.36%,误差率为2.70%。试验结果表明,此方法能够克服环境中的多种噪声干扰,能够在不同光照条件下对麦穗进行计数和产量预估。

Estimation method of wheat yield based on convolution neural network

Wheat yield is one of the important indices to evaluate agricultural productivity. In order to solve the problem of wheat yield prediction, convolution neural network was applied. It could provide reference for the estimation of agricultural productivity and guide the decision-making of agricultural production management. We constructed the wheat dataset from the images which were collected by unmanned aerial vehicle in Xinxiang and Luohe of Henan Province, and divided them into positive samples (wheat) and negative samples (leaf and background). The convolutional neural network recognition model was designed for normal physiological morphology and growth environment of wheat, and the image pyramid was used to construct multi-scale sliding windows. Finally, Non-maximum suppression (NMS) was used to remove the object frame with high overlap rate to realize counting of wheat ears per unit area. We randomly assigned 100 different wheat images to test. Compared with manual counting result, the accuracy rate reached 97.30%, the missed detection rate was 0.34%, the wrong detection rate was 2.36%, and the error rate was 2.70%. The experimental results showed that this method could overcome many kinds of noise interference in the environment, and could count wheat ear and estimate the yield of wheat accurately.