复杂背景下改进人工神经网络的储藏小麦中害虫图像识别  

Improved Artificial Neural Network for Image Recognition of Pests in Stored Wheat under Complex Backgrounds

作　　者：刘娜 陈章宝 王艳春 杨艳 LIU Na;CHEN Zhangbao;WANG Yanchun;YANG Yan(School of Electronic and Electrical Engineering,Bengbu University,Bengbu 233030,China)

机构地区：[1]蚌埠学院电子与电气工程学院,安徽蚌埠233030

出　　处：《信阳农林学院学报》2025年第1期98-102,110,共6页Journal of Xinyang Agriculture and Forestry University

基　　金：安徽省高校自然科学研究重点项目(2023AH052933,2023ZR02);蚌埠学院自然科学研究项目(2023ZR02)。

摘　　要：在小麦害虫图像检测识别中,因害虫的种类繁多,形态各异,且往往受到光照、背景等环境因素的影响,增加常规神经网络的训练难度,使得常规神经网络在训练过程中出现梯度消失的问题,从而导致最终识别结果的精度偏低。为此,对常规神经网络进行优化,研究并设计一种小麦害虫图像识别方法。将采集的小麦害虫彩色图像转换为灰度图像,利用小波变换技术去除图像噪声,结合形态学操作滤除图像的冗余信息,提高图像的质量。对滤波后的小麦害虫图像进行线性变换和对比度调整,实现对数据集的离线增广,构建出一个包含大量样本、类别丰富的小麦害虫图像数据集。针对数据集中的图像提取小麦害虫特征,引入神经网络算法构建识别模型,通过增加分支数与卷积层数,对模型进行改进与优化,以降低网络的深度。将小麦害虫特征输入至改进后的模型中实现小麦害虫图像识别。实验结果表明,本研究方法能够准确输出小麦害虫的图像类型,识别准确率较高。In wheat pest image detection and recognition,due to the wide variety and diverse forms of pests,as well as the influence of environmental factors such as lighting and background,the training difficulty of conventional neural networks is increased,resulting in the problem of gradient vanishing during the training process,which leads to low accuracy of the final recognition results.To this end,a wheat pest image recognition method is studied and designed by optimizing conventional neural networks.Convert the collected color images of wheat pests into grayscale images,use wavelet transform technology to remove image noise,and combine morphological operations to filter out redundant information in the images,improving the quality of the images.Perform linear transformation and contrast adjustment on the filtered wheat pest images to achieve offline augmentation of the dataset,and construct a wheat pest image dataset with a large number of samples and rich categories.To extract wheat pest features from images in the dataset,a neural network algorithm is introduced to construct a recognition model.By increasing the number of branches and convolution layers,the model is improved and optimized to reduce the depth of the network.The wheat pest features are input into the improved model to achieve wheat pest image recognition.The experimental results show that this research method can accurately output the image types of wheat pests with high recognition accuracy.

关 键 词：复杂背景 储藏小麦 害虫识别 图像处理 人工神经网络 

分 类 号：S126[农业科学—农业基础科学] TP391.41[自动化与计算机技术—计算机应用技术]