基于改进轻量化网络MobileViT的苹果叶片病虫害识别方法

针对苹果叶片病害识别准确率低以及现有模型难以适应真实复杂场景等问题,提出一种改进的轻量化网络——MobileViT_filter_FCN,以提高对苹果叶片病害的识别准确率,并使得模型可以适应户外的复杂光照及遮挡环境。首先收集5类常见苹果叶片病害(如落叶病、褐斑病等)的图像样本,并利用多种数据增强技术对样本数据进行预处理(如水平翻转、垂直翻转等),以增加样本数据的多样性并提高模型的泛化能力;接着利用傅里叶变换技术设计一个可学习的滤波器层Filter layer,替换原始MobileViT模型中的多头注意力结构,以降低图片中的噪声影响并提高模型性能;最后,在修改后的MobileViT模型基础上,利用深度卷积层和残差结构设计一种FCN结构,结合该结构增强模型对病害图像的特征学习能力,进一步提高模型性能。试验结果表明,改进后的MobileViT_filter模型对苹果叶片病害的平均识别准确率达到97.73%,较原模型提高0.95百分点;在该基础上加入FCN结构后,平均识别准确率达到98.03%,较原模型提高1.25百分点,同时参数量减少2.6 M。

Study on the Impact of Filter Layer Permeability on Revetment Top Layer Stability under Wave Action

In the design of revetment engineering under wave action, to resist the wave action, the pattern of top layer-filter layer-core （subsoil） is often adopted. In general, the structure of top layer is usually single discrete blocks, typically accropode blocks, four-leg square hollow blocks and barrier boards, and also acropode, riprap, paved rock blocks or concrete slabs with smaller waves. Such top layer has been provided with many research findings on its stability and is widely used in engineering. Setting a filter layer between the top layer and the lower dike core mainly has two functions： （1） giving certain permeability, to minimize the hydrodynamic load directly acting on the lower foundation soil; （2） giving certain hydraulic tightness, to prevent fine sediment of the lower foundation soil from being washed out. This paper is focused on a special filter layer with geotextile as its upper structure and coarse aggregate as its lower structure. By simulating geotextile with different permeability and coarse aggregate with different size, the pressure of top of cover layer and the down side of the geotextile is tested under wave actions, and compared with theoretical analysis, in this way, how the permeability of geotextile impacts the stability of top layer is studied. The research shows that when the filter layer under the geotextile has high permeability and the geotextile＇s permeability gets poorer, the uplift force to geotextile and the top layer will be increased under wave action, which will cause damage to the top layer when it is greater than the vertical component of the underwater gravity along the slope surface.

High-accuracy target tracking for multistatic passive radar based on a deep feedforward neural network

In radar systems,target tracking errors are mainly from motion models and nonlinear measurements.When we evaluate a tracking algorithm,its tracking accuracy is the main criterion.To improve the tracking accuracy,in this paper we formulate the tracking problem into a regression model from measurements to target states.A tracking algorithm based on a modified deep feedforward neural network(MDFNN)is then proposed.In MDFNN,a filter layer is introduced to describe the temporal sequence relationship of the input measurement sequence,and the optimal measurement sequence size is analyzed.Simulations and field experimental data of the passive radar show that the accuracy of the proposed algorithm is better than those of extended Kalman filter(EKF),unscented Kalman filter(UKF),and recurrent neural network(RNN)based tracking methods under the considered scenarios.

Application of Biochar on the runoff purification performance of porous asphalt pavement

Biochar is currently applied in many low-impact development measures,such as biofilters and filter strips.However,its application in permeable pavement is limited.Due to the wide range of raw material sources,biochar may also contain nutrients and heavy metals.Whether its leachate will cause contamination during application remains unknown.Based on the static leachate contamination test,this study first evaluated the leachate contamination risks of three types of biochar and porous asphalt mixture(PA)with three biochar fillers.Through the pavement infiltration test,the purification performance of PA with biochar fillers and filter layers was further analysed.The results show that biochar contains nitrogen and phosphorus,and the leaching contamination of coconut shell is the most obvious;when the biochar is applied as a filler in PA,the leaching contamination decreases,while the pollutant purification performance is not obvious;when 3−5 mm rice straw biochar is used as a filter layer,the removal rate of total suspended solids can be 100%,but it is accompanied by more serious nitrogen and phosphorus leaching,and the greater the thickness,the more serious the degree of leaching.Deionized water cleaning can remove the phosphates contained in biochar,but have little effect on nitrogen.Therefore,before the application of biochar,cleaning measures should be taken to minimize its leaching contamination.

Tunable terahertz filter based on alternative liquid crystal layers and metallic slats

We propose and demonstrate a pseudo Fabry–Pérot filter in the terahertz frequency range of 0.1–0.5 THz. It consists of alternative liquid crystal layers and metallic slats. Separate sharp resonant peaks are shown in the simulated transmission spectra, and their positions shift toward higher frequencies when the refractive index of liquid crystal decreases. The measured transmission spectra are consistent with corresponding simulations. Via thermally tuning the refractive index of the filled liquid crystal, the resonant transmission frequencies shift accordingly. The work supplies a novel design for tunable terahertz filters, which would play important roles in terahertz imaging, sensing, high speed communication, and security applications.

Optimization of a guided-mode resonance filter by varying the thickness of the buffer layer

In this Letter, we demonstrate that by adjusting the thickness of the buffer layer, the optical responses of a guided-mode resonance filter （GMRF） can be improved for sensor applications. The GMRF is fabricated using a replica molding with a plastic substrate and a UV-curable polymer. SiO2 buffer layers of different thicknesses are deposited before the waveguide-layer deposition. The sensitivity of the GMRFs decreases slightly with in- creasing SiO2 layer thickness. By contrast, the full width at half-maximum reduces substantially with increasing SiO2 layer thickness, resulting in the improvement of the overall figure of merit.