Objective Intelligent detection of foxtail millet blast disease in field environments faces challenges such as strong background interference, varying lesion scales, and limited adaptability of existing models in real-world applications. Methods To address these issues, this study leveraged a large-scale foxtail millet blast dataset and proposed an improved detection model, YOLOv8-SDL,based on YOLOv8n. Results The model enhanced detection stability and accuracy in complex backgrounds through three key modifications: strengthening backbone network feature extraction, introducing lightweight and efficient upsampling to improve feature fusion, and incorporating an attention mechanism to enhance critical feature selection. First, the Switchable Atrous Convolution (SAC)structure was optimized by adding a 3×3 convolution layer before the 1×1 convolution in its context structure. Combined with the Squeeze-and-Excitation (SE) channel attention mechanism, the SE-SAC module was constructed and embedded into the C2f module of the backbone network to enhance multi-scale feature extraction of lesion areas. Second,DySample upsampling replaced nearest-neighbor interpolation. By employing a dynamic point-sampling strategy,DySample reduced computational overhead while minimizing feature information loss during upsampling, thereby improving feature fusion and small lesion localization accuracy. Third, the Large Separable Kernel Attention (LSKA) mechanism was integrated into the neck network. Its separable convolution design enhanced the model's ability to focus on disease lesion features in complex backgrounds. Experimental results showed that YOLOv8-SDL achieved a detection accuracy of 91.5%, mAP@0.5 of 93.9%, and mAP@0.5~0.95 of 62.0%, outperforming the original model by 4.4%, 1.4%, and 2.2%, respectively. Conclusion This model provided robust and reliable technical support for accurate foxtail millet blast detection in challenging field environments.
谷子(Setaria italica L.),一年生,种子脱皮后俗称小米,是我国北方传统的粮食作物,因其耐旱、耐瘠、适应性强等特点,是发展有机旱作农业的重要作物[1]。近年来由于缺少抗病品种和单一品种大规模种植,谷瘟病已成为我国谷子生产的主要病害之一[2]。谷瘟病是由灰梨孢菌引起的一种病害,严重时导致叶片枯黄、茎秆倾斜倒伏甚至枯死,对谷子产量和品质造成显著威胁[3]。由于谷瘟病病斑尺度多变,传统的检测方法难以精准识别,还易出现漏检、误检,因此对谷瘟病的智能化快速检测和精准防治开展研究,对减少谷子产量损失、保障其稳产增产具有重要的现实意义。
在模型的性能评价上,本研究选用了浮点量、参数量、模型大小作为主要评价依据。浮点量即浮点运算量(Floating point operations per second,FLOPs),用来衡量模型在训练时运算的复杂度,反映了模型对硬件运算能力的要求。参数量(Parameters)表示模型的复杂程度,模型结构越简单则参数量越少。模型大小指的是训练完成后模型所占存储空间的大小,它直接影响到模型的可部署性和运行效率。评价模型性能时必须综合考虑上述各指标。
LiG Y, BaiX Y, YinC K, et al. Occurrence status and pathogen identification of millet nematode disease in Shandong Province[J].Shandong Agricultural Sciences,2024,56(3):125-131.
LvP K, SuH Z, LüC,et al. An illustrated book of primary colors of diseases and pests of Chinese food crops, economic erops, and medicinal plants[M]. Hohhot:Yuanfang Publishing House,1999:169-170.
[7]
SunH N, XuH W, LiuB, et al. MEAN-SSD: a novel real-time detector for apple leaf diseases using improved light-weight convolutional neural networks[J]. Computers and Electronics in Agriculture, 2021, 189: 106379.
[8]
RedmonJ, DivvalaS, GirshickR,et al.You only look once: unified, real-time object detection[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas:IEEE, 2016: 779-788.
ZhangL X, JingJ P, LiS F, et al. Tomato disease recognition system based on image automatic labeling and improved YOLOv5[J]. Transactions of the Chinese Society for Agricultural Machinery,2023,54(11):198-207.
YueK, ZhangP C, WangL, et al. Recognizing citrus in complex environment using improved YOLOv8n[J]. Transactions of the Chinese Society of Agricultural Engineering,2024,40(8):152-158.
[13]
DongC, ZhangK, XieZ Y, et al.An improved cascade RCNN detection method for key components and defects of transmission lines[J].IET Generation, Transmission & Distribution,2023,17(19):4277-4292.
[14]
WangX L, ShrivastavaA, GuptaA. A-fast-RCNN: hard positive generation via adversary for object detection[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR),Honolulu:IEEE,2017: 3039-3048.
[15]
RenS Q, HeK M, GirshickR, et al. Faster R-CNN:towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,39(6):1137-1149.
MaoR, ZhangY C, WangZ X, et al. Recognizing stripe rust and yellow dwarf of wheat using improved Faster-RCNN[J]. Transactions of the Chinese Society of Agricultural Engineering,2022,38(17):176-185.
YanJ W, ZhaoY, ZhangL W,et al. Recognition of Rosa roxbunghii fruit in natural environment based on residual network[J]. Journal of Chinese Agricultural Mechanization,2020,41(10):191-196.
ZhangH T, LiY J, TanL, et al. Image recognition of millet leaf disease based on CS-SVM[J]. Acta Agriculturae Zhejiangensis,2020,32(2):274-282.
[22]
李艺嘉.基于机器视觉的谷子叶片病害的图像识别[D].郑州:华北水利水电大学,2020.
[23]
LiY J. Image recognition of millet leaf disease based on machine vision[D].Zhengzhou:North China University of Water Resources and Electric Power,2020.
ZhangH T, LuoY M, TanL, et al. Research on millet disease identification based on transfer learning and residual network[J].Journal of Henan Agricultural Sciences,2023,52(12):162-171.
[26]
LiuS, QiL, QinH, et al. Path aggregation network for instance segmentation[C].2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City:IEEE, 2018: 8759-8768.
[27]
LinT Y, DollárP, GirshickR, et al. Feature pyramid networks for object detection[C].2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR),Honolulu:IEEE,2017:936-944.
[28]
QiaoS, ChenL C, YuilleA. DetectoRS: detecting objects with recursive feature pyramid and switchable atrous convolution[C].2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),Nashville:IEEE,2021: 10208-10219.
[29]
HuJ, ShenL, SunG. Squeeze-and-excitation networks[C].2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City:IEEE, 2018:7132-7141.
[30]
LiuW Z, LuH, FuH T, et al. Learning to upsample by learning to sample[C]. 2023 IEEE/CVF International Conference on Computer Vision (ICCV), Paris: IEEE, 2023, 6004-6014.
[31]
LauK W, PoL M, RehmanY A U. Large separable kernel attention: rethinking the large kernel attention design in CNN[J]. Expert Systems with Applications,2024,236:121352.
[32]
WangJ, ChenK, XuR, et al. CARAFE: Content-Aware ReAssembly of FEatures[C]. 2019 IEEE/CVF International Conference on Computer Vision (ICCV), Seoul: IEEE, 2019: 3007-3016.
[33]
LuH, LiuW Z, FuH, et al. FADE: fusing the assets of decoder and encoder for task agnostic upsampling[C]//Avidan S, Brostow G, Cissé M, et al. Computer Vision–ECCV 2022, Cham: Springer, 2022: 231-247.