Recognition of weeds at asparagus fields using multi-feature fusion and backpropagation neural network
Recognition of weeds at asparagus fields using multi-feature fusion and backpropagation neural network
复制标题
使用多特征融合和反向传播神经网络识别芦笋田杂草
DOI：
10.25165/j.ijabe.20211404.6135
复制
发表时间：
2021-07-01
影响因子：
2.4
通讯作者：
Mao, Hanping
中科院分区：
文献类型：
--
作者：
Wang, Yafei;Zhang, Xiaodong;Mao, Hanping
为了解决单一特征对杂草识别率低的问题，提出了一种基于特征的芦笋（Asparagus officinalis L.）领域使用多特征融合和反向传播神经网络（BPNN）。共收集到382幅杂草与芦笋生长竞争的图像，其中包括135幅小蓟（Cirsiumarvense（L.）斯科普，138份，其中Conyza sumatrensis（Retz.）E.步行者和常春藤（Calystegia hederacea Wall.利用2G-R-B因子从杂草的RGB图像中提取灰度图像。采用大津法对杂草灰度图像进行阈值分割。然后通过膨胀和腐蚀形态学操作对叶片内部孔洞进行填充，并去除其他干扰目标，得到二值图像。通过对二值图像和RGB图像进行掩模处理得到前景图像。然后，采用颜色矩算法提取杂草颜色特征，采用灰度共生矩阵和局部二值模式（LBP）算法提取杂草纹理特征，提取7个Hu不变矩特征和杂草的圆度、细长度作为杂草的形状特征。根据测试样本的形状、颜色、纹理和融合特征，建立杂草识别模型。试验结果表明，小蓟的识别率为100%~ 100%;斯科普，常春藤和Conyza sumatrensis（Retz.）E.步行者识别率分别为82.72%（颜色特征）、72.41%（形状特征）、86.73%（纹理特征）和93.51%（融合特征）。该方法可为芦笋田杂草识别研究提供参考。
In order to solve the problem of low recognition rates of weeds by a single feature, a method was proposed in this study to identify weeds in Asparagus (Asparagus officinalis L.) field using multi-feature fusion and backpropagation neural network (BPNN). A total of 382 images of weeds competing with asparagus growth were collected, including 135 of Cirsium arvense (L.) Scop., 138 of Conyza sumatrensis (Retz.) E. Walker, and 109 of Calystegia hederacea Wall. The grayscale images were extracted from the RGB images of weeds using the 2G-R-B factor. Threshold segmentation of the grayscale image of weeds was applied using Otsu method. Then the internal holes of the leaves were filled through the expansion and corrosion morphological operations, and other interference targets were removed to obtain the binary image. The foreground image was obtained by masking the binary image and the RGB image. Then, the color moment algorithm was used to extract weeds color feature, the gray level co-occurrence matrix and the Local Binary Pattern (LBP) algorithm was used to extract weeds texture features, and seven Hu invariant moment features and the roundness and slenderness ratio of weeds were extracted as their shape features. According to the shape, color, texture, and fusion features of the test samples, a weed identification model was built. The test results showed that the recognition rate of Cirsium arvense (L.) Scop., Calystegia hederacea Wall. and Conyza sumatrensis (Retz.) E. Walker were 82.72% (color feature), 72.41% (shape feature), 86.73% (texture feature) and 93.51% (fusion feature), respectively. Therefore, this method can provide a reference for the study of weeds identification in the asparagus field.