[1]齐国红,许新华,师晓丽.基于多尺度注意力U-Net的结球甘蓝青虫检测方法[J].江苏农业学报,2023,(06):1349-1357.[doi:doi:10.3969/j.issn.1000-4440.2023.06.010]
　QI Guo-hong,XU Xin-hua,SHI Xiao-li.Detection of cabbage worms based on multi-scale attention U-Net[J].,2023,(06):1349-1357.[doi:doi:10.3969/j.issn.1000-4440.2023.06.010]
点击复制

基于多尺度注意力U-Net的结球甘蓝青虫检测方法()
分享到：

江苏农业学报[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2023年06期
页码:
1349-1357
栏目:
农业信息工程
出版日期:
2023-09-30

文章信息/Info

Title:
Detection of cabbage worms based on multi-scale attention U-Net
作者:
齐国红; 许新华; 师晓丽
(郑州西亚斯学院电子信息工程学院，河南郑州451150)
Author(s):
QI Guo-hong; XU Xin-hua; SHI Xiao-li
(School of Electronic Information Engineering, Zhengzhou SIAS University, Zhengzhou 451150, China)
关键词:
结球甘蓝青虫检测; 超像素聚类; U-Net; 多尺度注意力U-Net
Keywords:
cabbageworm detection; superpixel clustering; U-Net; multiscale attention U-Net
分类号:
TP391.41; S432
DOI:
doi:10.3969/j.issn.1000-4440.2023.06.010
文献标志码:
A
摘要:
针对结球甘蓝青虫姿态多样、形状不规则以及传统U-Net对多尺度图像检测的鲁棒性较差等问题，本研究提出一种基于多尺度注意力U-Net(MSAU-Net)的结球甘蓝青虫检测方法。该方法将多尺度空洞Inception和注意力引入到U-Net，通过设置不同膨胀率的初始卷积层卷积核和全局池化层类型，提取多尺度深层次的结球甘蓝青虫检测特征。首先，对原始图像进行超像素聚类，极大减少结球甘蓝青虫图像的基元数量；其次，利用多尺度空洞U-Net提取不同大小的结球甘蓝青虫特征；最后，通过注意力连接将MSAU-Net同层的浅层、深层特征拼接，得到结球甘蓝青虫图像的关键特征，加快网络训练。MSAU-Net方法在结球甘蓝青虫数据集上的平均检测精度为95.26%，较U-Net方法提高了约6个百分点。MSAU-Net方法能较好地检测到大小不同的结球甘蓝青虫，能够应用于结球甘蓝青虫自动检测系统。
Abstract:
Aiming at the problems of diverse postures, irregular shapes, and the low robustness of traditional U-Net to multi-scale image detection, a detection method of cabbage worms based on multi-scale attention U-Net (MSAU-Net) was proposed. This method introduced multi-scale dilated Inception and attention into U-Net, and extracted multi-scale high-level worm detection features by setting initial convolution layer convolution core and global pooling layer types with different expansion rates. First, the original image was clustered by superpixel cluster algorithm to greatly reduce the number of primitives of worm images. Then, the multi-scale dilated U-Net was used to extract the features of different sizes of worms. Finally, the shallow and deep features of the same layer of MSAU-Net were spliced through attention connection to obtain the key features of worm images and speed up network training. The average detection accuracy on the cabbageworm dataset was 95.26%, which was about six percentage points higher than that of U-Net. The MSAU-Net method can detect cabbageworms with different sizes, and can be applied to the automatic detection system of cabbage worms.

参考文献/References:

[1]康飞龙,李佳,刘涛,等. 多类农作物病虫害的图像识别应用技术研究综述[J]. 江苏农业科学, 2020,48(22):22-27.
[2]张善文,邵彧,齐国红,等. 基于多尺度注意力卷积网络的作物害虫检测[J]. 江苏农业学报,2021,37(3):579-588.
[3]YAAKOB S N, JAIN L. An insect classification analysis based on shape features using quality threshold ARTMAP and moment invariant[J]. Applied Intelligence, 2012, 37:12-30.
[4]MARTINEAU M, CONTE D, RAVEAUX R, et al. A survey on image-based insect classification [J]. Pattern Recognition, 2017, 65:273-284.
[5]EBRAHIMI M A, KHOSHTAGHAZA M H, MINAEI S, et al. Vision-based pest detection based on SVM classification method[J]. Computers and Electronics in Agriculture, 2017, 137:52-58.
[6]RAWAT W, WANG Z. Deep convolutional neural networks for image classification: a comprehensive review[J]. Neural Computation, 2017, 29(9): 2352-2449.
[7]LI Y, WANG H, DANG L M, et al. Crop pest recognition in natural scenes using convolutional neural networks[J]. Computers and Electronics in Agriculture, 2020, 169. DOI:10.1016/j.compag.2019.105174.
[8]XIA D, CHEN P, WANG B, et al. Insect detection and classification based on an improved convolutional neural network[J]. Sensors, 2018, 18(12). DOI: 10.3390/s18124169.
[9]LIU L, WANG R, XIE C, et al. PestNet: an end-to-end deep learning approach for large-scale multi-class pest detection and classification[J]. IEEE Access, 2019, 7: 45301-45312.
[10]WANG Q J, ZHANG S Y, DONG S F, et al. Pest24: a large-scale very small object data set of agricultural pests for multi-target detection[J]. Computers and Electronics in Agriculture, 2020, 175. DOI:10.1016/j.compag.2020.105585.
[11]张博,张苗辉,陈运忠. 基于空间金字塔池化和深度卷积神经网络的作物害虫识别[J].农业工程学报, 2019, 35(19):209-215.
[12]LIU X, ZHANG Y, JING H, et al. Ore image segmentation method using U-Net and Res-Unet convolutional networks[J]. RSC Advances, 2020, 10(16): 9396-9406.
[13]SEVASTOPOLSKY A. Optic disc and cup segmentation methods for glaucoma detection with modification of U-Net convolutional neural network[J]. Pattern Recognition and Image Analysis, 2017, 27(3): 618-624.
[14]孙俊,何小飞,谭文军,等. 空洞卷积结合全局池化的卷积神经网络识别作物幼苗与杂草[J].农业工程学报,2018,34(11):159-165.
[15]YANG X. An overview of the attention mechanisms in computer vision[C]. Inner Mongolia:IOP Publishing, 2020.
[16]LI A, CHEN J, KANG B, et al. Adaptive multi-attention convolutional neural network for fine-grained image recognition[C]. Waikoloa: IEEE, 2019.
[17]WANG M, LIU X, GAO Y, et al. Superpixel segmentation: a benchmark[J]. Signal Processing: Image Communication, 2017, 56: 28-39.
[18]QIN F, GUO J, LANG F. Superpixel segmentation for polarimetric SAR imagery using local iterative clustering[J]. IEEE Geoscience and Remote Sensing Letters, 2015, 12(1): 13-17.
[19]ACHANTA R, SHAJI A, SMITH K, et al. SLIC superpixels compared to state-of-the-art superpixel methods[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(11): 2274-2282.
[20]YANG F, SUN Q, JIN H, et al. Superpixel segmentation with fully convolutional networks[C]. Virtual: IEEE, 2020.
[21]LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[C]. Boston: IEEE, 2015.

备注/Memo

备注/Memo:
收稿日期:2022-09-17基金项目:国家自然科学基金项目(62172338)；河南省科技厅项目(232102110274、222102210122)；河南省高等学校重点科研项目(23B510004、22B520049)；河南省教育厅2022年民办普通高等学校学科专业建设资助项目[教办政法(2022)219号]作者简介:齐国红(1987-),女，河南郑州人，硕士，讲师，主要从事模糊模式识别及其应用研究。(E-mail)919728600@qq.com
更新日期/Last Update: 2023-11-17