中国农机化学报2026，Vol.47Issue(2)：202-209,8.DOI:10.13733/j.jcam.issn.2095-5553.2026.02.027
基于改进DenseNet的棉叶螨危害等级识别研究
Research on cotton leaf mite hazard level recognition based on improved DenseNet
摘要
Abstract
In response to the labor-intensive and time-consuming nature,as well as the lagging issues inherent in traditional manual diagnosis and grading methods for cotton leaf mites,a novel cotton leaf mite hazard level recognition model based on an improved DenseNet—121 is proposed.Following the standard classification criteria for cotton leaf mite damage,images of cotton leaves with varying hazard levels under both uniform and natural backgrounds were collected.The dataset was augmented to simulate the influence of diverse weather conditions,shooting angles,and device noise during image acquisition.Considering the high similarity between different mite damage levels and the resulting identification challenges,we implemented three key improvements to the DenseNet—121 model.First,the 7×7 convolution kernel in the initial convolution layer was replaced with an Inception module to enhance the feature extraction capability of the shallow network layers.Second,a SimAM attention mechanism was introduced after the Transition Layer to emphasize cotton leaf mite hazard features and suppress background features.Lastly,DropBlock regularization was applied after the Dense Layer to enhance the model's robustness and prevent overfitting.The results demonstrate that the original model achieves a recognition accuracy of 90.76%on the augmented dataset,representing an improvement of 4.21 percentage points over the original model.Individually,data augmentation and the three improvement strategies increase the recognition accuracy of the model by 1.47,2.74,2.37 and 1.86 percentage points,respectively.The comprehensive performance of the model is notably superior to that of other models such as VGG16 and ResNet50.关键词
棉叶/螨害分级/改进DenseNet—121/注意力机制/正则化Key words
cotton leaf/mite hazard level/improved DenseNet—121/attention mechanism/regularization分类
农业科技引用本文复制引用
Lei Junjie,Zhou Baoping,Zhang Yan..基于改进DenseNet的棉叶螨危害等级识别研究[J].中国农机化学报,2026,47(2):202-209,8.基金项目
国家自然科学基金项目(61563046) （61563046）
塔里木大学研究生科研创新项目(TDGRI202358) （TDGRI202358）
