融合SKNet与MobilenetV3的芒果叶片病虫害分类方法

【目的】针对芒果叶片病虫害缺少数据集和识别准确率低的问题,筛选构建芒果叶片病虫害分类模型,以提高芒果叶病虫害分类准确率。【方法】提出使用去噪扩散模型进行病虫害数据增强,同时联合SKNet与MobilenetV3模型的芒果叶片病虫害分类方法。首先使用去噪扩散模型对数据集进行扩充,再采用多尺度结构相似性指标对生成的病虫害图像与拍摄的病虫害图像之间的相似程度进行评估,接着对DDIM与DCGAN网络训练和生成效果进行比对。在MobilenetV3模型中,将SE注意力模块替换为SKNet模块进行构建网络模型。【结果】使用DDIM生成的所有类型的病虫害图像与拍摄的病虫害图像的MS-SSIM指标均大于0.63,且都高于DCGAN。相较于其他注意力模块,联合SKNet与MobilenetV3的分类效果最佳,在98%以上。对添加CA、CBAM、ECA注意力模块进行平滑类激活图可视化,对比其他注意力模块,使用SKNet注意力分布区域更为集中在病虫害叶片上。【结论】该方法在病虫害叶片检测上具有良好的应用前景,能提升病虫害识别效率与精度,减少检测成本,同时可应用于移动式或者嵌入式设备。

Deep neural network based on multi-level wavelet and attention for structured illumination microscopy

Structured illumination microscopy(SIM)is a popular and powerful super-resolution(SR)technique in biomedical research.However,the conventional reconstruction algorithm for SIM heavily relies on the accurate prior knowledge of illumination patterns and signal-to-noise ratio(SNR)of raw images.To obtain high-quality SR images,several raw images need to be captured under high fluorescence level,which further restricts SIM’s temporal resolution and its applications.Deep learning(DL)is a data-driven technology that has been used to expand the limits of optical microscopy.In this study,we propose a deep neural network based on multi-level wavelet and attention mechanism(MWAM)for SIM.Our results show that the MWAM network can extract high-frequency information contained in SIM raw images and accurately integrate it into the output image,resulting in superior SR images compared to those generated using wide-field images as input data.We also demonstrate that the number of SIM raw images can be reduced to three,with one image in each illumination orientation,to achieve the optimal tradeoff between temporal and spatial resolution.Furthermore,our MWAM network exhibits superior reconstruction ability on low-SNR images compared to conventional SIM algorithms.We have also analyzed the adaptability of this network on other biological samples and successfully applied the pretrained model to other SIM systems.

Joint Rain Streaks & Haze Removal Network for Object Detection

In the realm of low-level vision tasks,such as image deraining and dehazing,restoring images distorted by adverse weather conditions remains a significant challenge.The emergence of abundant computational resources has driven the dominance of deep Convolutional Neural Networks(CNNs),supplanting traditional methods reliant on prior knowledge.However,the evolution of CNN architectures has tended towards increasing complexity,utilizing intricate structures to enhance performance,often at the expense of computational efficiency.In response,we propose the Selective Kernel Dense Residual M-shaped Network(SKDRMNet),a flexible solution adept at balancing computational efficiency with network accuracy.A key innovation is the incorporation of an M-shaped hierarchical structure,derived from the U-Net framework as M-Network(M-Net),within which the Selective Kernel Dense Residual Module(SDRM)is introduced to reinforce multi-scale semantic feature maps.Our methodology employs two sampling techniques-bilinear and pixel unshuffled and utilizes a multi-scale feature fusion approach to distil more robust spatial feature map information.During the reconstruction phase,feature maps of varying resolutions are seamlessly integrated,and the extracted features are effectively merged using the Selective Kernel Fusion Module(SKFM).Empirical results demonstrate the comprehensive superiority of SKDRMNet across both synthetic and real rain and haze datasets.

基于SKNet的增强型Pix2pixHD图像去雾方法

针对采用去雾算法处理后图片仍存在色彩失真、残雾遗留等问题,提出一种基于SKNet的增强型Pix2pixHD去雾方法,即在增强型Pix2pixHD网络的增强模块中引入SKNet网络,实现了不同尺度特征的选择和融合,提高了算法对图像特征的利用率。此外,在SKNet网络前添加可变形卷积,使得卷积核产生自由形变,以适应不规则的目标物体,增强了模型的可变换能力,达到了很好的去雾效果。较现有的去雾方法,文章提出的网络更好地利用了图像各个尺度的特征细节,提高了图像去雾的效果。

Selective ensemble modeling based on nonlinear frequency spectral feature extraction for predicting load parameter in ball mills

Strong mechanical vibration and acoustical signals of grinding process contain useful information related to load parameters in ball mills. It is a challenge to extract latent features and construct soft sensor model with high dimensional frequency spectra of these signals. This paper aims to develop a selective ensemble modeling approach based on nonlinear latent frequency spectral feature extraction for accurate measurement of material to ball volume ratio. Latent features are first extracted from different vibrations and acoustic spectral segments by kernel partial least squares. Algorithms of bootstrap and least squares support vector machines are employed to produce candidate sub-models using these latent features as inputs. Ensemble sub-models are selected based on genetic algorithm optimization toolbox. Partial least squares regression is used to combine these sub-models to eliminate collinearity among their prediction outputs. Results indicate that the proposed modeling approach has better prediction performance than previous ones.

A selective kernel-based cycle-consistent generative adversarial network for unpaired low-dose CT denoising

Low-dose computed tomography(LDCT)denoising is an indispensable procedure in the medical imaging field,which not only improves image quality,but can mitigate the potential hazard to patients caused by routine doses.Despite the improvement in performance of the cycle-consistent generative adversarial network(CycleGAN)due to the well-paired CT images shortage,there is still a need to further reduce image noise while retaining detailed features.Inspired by the residual encoder–decoder convolutional neural network(RED-CNN)and U-Net,we propose a novel unsupervised model using CycleGAN for LDCT imaging,which injects a two-sided network into selective kernel networks(SK-NET)to adaptively select features,and uses the patchGAN discriminator to generate CT images with more detail maintenance,aided by added perceptual loss.Based on patch-based training,the experimental results demonstrated that the proposed SKFCycleGAN outperforms competing methods in both a clinical dataset and the Mayo dataset.The main advantages of our method lie in noise suppression and edge preservation.