GNN Representation Learning and Multi-Objective Variable Neighborhood Search Algorithm for Wind Farm Layout Optimization

With the increasing demand for electrical services,wind farm layout optimization has been one of the biggest challenges that we have to deal with.Despite the promising performance of the heuristic algorithm on the route network design problem,the expressive capability and search performance of the algorithm on multi-objective problems remain unexplored.In this paper,the wind farm layout optimization problem is defined.Then,a multi-objective algorithm based on Graph Neural Network(GNN)and Variable Neighborhood Search(VNS)algorithm is proposed.GNN provides the basis representations for the following search algorithm so that the expressiveness and search accuracy of the algorithm can be improved.The multi-objective VNS algorithm is put forward by combining it with the multi-objective optimization algorithm to solve the problem with multiple objectives.The proposed algorithm is applied to the 18-node simulation example to evaluate the feasibility and practicality of the developed optimization strategy.The experiment on the simulation example shows that the proposed algorithm yields a reduction of 6.1% in Point of Common Coupling(PCC)over the current state-of-the-art algorithm,which means that the proposed algorithm designs a layout that improves the quality of the power supply by 6.1%at the same cost.The ablation experiments show that the proposed algorithm improves the power quality by more than 8.6% and 7.8% compared to both the original VNS algorithm and the multi-objective VNS algorithm.

Computer Vision and Deep Learning-enabled Weed Detection Model for Precision Agriculture

Presently,precision agriculture processes like plant disease,crop yield prediction,species recognition,weed detection,and irrigation can be accom-plished by the use of computer vision(CV)approaches.Weed plays a vital role in influencing crop productivity.The wastage and pollution of farmland's natural atmosphere instigated by full coverage chemical herbicide spraying are increased.Since the proper identification of weeds from crops helps to reduce the usage of herbicide and improve productivity,this study presents a novel computer vision and deep learning based weed detection and classification(CVDL-WDC)model for precision agriculture.The proposed CVDL-WDC technique intends to prop-erly discriminate the plants as well as weeds.The proposed CVDL-WDC technique involves two processes namely multiscale Faster RCNN based object detection and optimal extreme learning machine(ELM)based weed classification.The parameters of the ELM model are optimally adjusted by the use of farmland fertility optimization(FFO)algorithm.A comprehensive simulation analysis of the CVDL-WDC technique against benchmark dataset reported the enhanced out-comes over its recent approaches interms of several measures.

Increasing Crop Quality and Yield with a Machine Learning-Based Crop Monitoring System

Farming is cultivating the soil,producing crops,and keeping livestock.The agricultural sector plays a crucial role in a country’s economic growth.This research proposes a two-stage machine learning framework for agriculture to improve efficiency and increase crop yield.In the first stage,machine learning algorithms generate data for extensive and far-flung agricultural areas and forecast crops.The recommended crops are based on various factors such as weather conditions,soil analysis,and the amount of fertilizers and pesticides required.In the second stage,a transfer learningbased model for plant seedlings,pests,and plant leaf disease datasets is used to detect weeds,pesticides,and diseases in the crop.The proposed model achieved an average accuracy of 95%,97%,and 98% in plant seedlings,pests,and plant leaf disease detection,respectively.The system can help farmers pinpoint the precise measures required at the right time to increase yields.

Real-Time Crop Prediction Based on Soil Fertility and Weather Forecast Using IoT and a Machine Learning Algorithm

The aim of this article is to assist farmers in making better crop selection decisions based on soil fertility and weather forecast through the use of IoT and AI (smart farming). To accomplish this, a prototype was developed capable of predicting the best suitable crop for a specific plot of land based on soil fertility and making recommendations based on weather forecast. Random Forest machine learning algorithm was used and trained with Jupyter in the Anaconda framework to achieve an accuracy of about 99%. Based on this process, IoT with the Message Queuing Telemetry Transport (MQTT) protocol, a machine learning algorithm, based on Random Forest, and weather forecast API for crop prediction and recommendations were used. The prototype accepts nitrogen, phosphorus, potassium, humidity, temperature and pH as input parameters from the IoT sensors, as well as the weather API for data forecasting. The approach was tested in a suburban area of Yaounde (Cameroon). Taking into account future meteorological parameters (rainfall, wind and temperature) in this project produced better recommendations and therefore better crop selection. All necessary results can be accessed from anywhere and at any time using the IoT system via a web browser.

A Chaotic Local Search-Based Particle Swarm Optimizer for Large-Scale Complex Wind Farm Layout Optimization

Wind energy has been widely applied in power generation to alleviate climate problems.The wind turbine layout of a wind farm is a primary factor of impacting power conversion efficiency due to the wake effect that reduces the power outputs of wind turbines located in downstream.Wind farm layout optimization(WFLO)aims to reduce the wake effect for maximizing the power outputs of the wind farm.Nevertheless,the wake effect among wind turbines increases significantly as the number of wind turbines increases in the wind farm,which severely affect power conversion efficiency.Conventional heuristic algorithms suffer from issues of low solution quality and local optimum for large-scale WFLO under complex wind scenarios.Thus,a chaotic local search-based genetic learning particle swarm optimizer(CGPSO)is proposed to optimize large-scale WFLO problems.CGPSO is tested on four larger-scale wind farms under four complex wind scenarios and compares with eight state-of-the-art algorithms.The experiment results indicate that CGPSO significantly outperforms its competitors in terms of performance,stability,and robustness.To be specific,a success and failure memories-based selection is proposed to choose a chaotic map for chaotic search local.It improves the solution quality.The parameter and search pattern of chaotic local search are also analyzed for WFLO problems.

Hybrid Deep Learning Method for Diagnosis of Cucurbita Leaf Diseases

In agricultural engineering,the main challenge is on methodologies used for disease detection.The manual methods depend on the experience of the personal.Due to large variation in environmental condition,disease diagnosis and classification becomes a challenging task.Apart from the disease,the leaves are affected by climate changes which is hard for the image processing method to discriminate the disease from the other background.In Cucurbita gourd family,the disease severity examination of leaf samples through computer vision,and deep learning methodologies have gained popularity in recent years.In this paper,a hybrid method based on Convolutional Neural Network(CNN)is proposed for automatic pumpkin leaf image classification.The Proposed Denoising and deep Convolutional Neural Network(CNN)method enhances the Pumpkin Leaf Pre-processing and diagnosis.Real time data base was used for training and testing of the proposed work.Investigation on existing pre-trained network Alexnet and googlenet was investigated is done to evaluate the performance of the pro-posed method.The system and computer simulations were performed using Matlab tool.

Effect of Panax notoginseng saponins on the expression of beta-amyloid protein in the cortex of the parietal lobe and hippocampus, and spatial learning and memory in a mouse model of senile dementia

BACKGROUND： The pharmacological actions of Panax notoginseng saponins （PNS） lie in removing free radicals, anti-inflammation and anti-oxygenation. It can also improve memory and behavior in rat models of Alzheimer＇s disease. OBJECTIVE： Using the Morris water maze, immunohistochemistry, real-time PCR and RT-PCR, this study aimed to measure improvement in spatial learning, memory, expression of amyloid precursor protein （App） and β -amyloid （A β ）, to investigate the mechanism of action of PNS in the treatment of AD in the senescence accelerated mouse-prone 8 （SAMP8） and compare the effects with huperzine A. DESIGN, TIME AND SETTING： A completely randomized grouping design, controlled animal experiment was performed in the Center for Research ＆ Development of New Drugs, Guangxi Traditional Chinese Medical University from July 2005 to April 2007. MATERIALS： Sixty male SAMP8 mice, aged 3 months, purchased from Tianjin Chinese Traditional Medical University of China, were divided into four groups： PNS high-dosage group, PNS low-dosage group, huperzine A group and control group. PNS was provided by Weihe Pharmaceutical Co., Ltd. （batch No.： Z53021485, Yuxi, Yunan Province, China）. Huperzine A was provided by Zhenyuan Pharmaceutical Co., Ltd. （batch No.： 20040801, Zhejiang, China）. METHODS： The high-dosage group and low-dosage group were treated with 93.50 and 23.38 mg/kg PNS respectively per day and the huperzine A group was treated with 0.038 6 mg/kg huperzine A per day, all by intragastric administration, for 8 consecutive weeks. The same volume of double distilled water was given to the control group. MAIN OUTCOME MEASURES： After drug administration, learning and memory abilities were assessed by place navigation and spatial probe tests. The recording indices consisted of escape latency （time-to-platform）, and the percentage of swimming time spent in each quadrant. The number of A β 1-40, A β 1-42 and App immunopositive neurons in the brains of SAMP8 mice was analyzed by immunohistochemistry. The mRNA content ofApp, tau, acetylcholinesterase, and synaptophysin （Syp） was tested by real time PCR and RT-PCR. RESULTS： The PCR results show that PNS can downregulate the expression of the App gene and upregulate the expression of the Syp gene in the parietal cortex and hippocampus of SAMP8 mice. The therapeutic effects of the PNS high-dosage group were greater than those of the PNS low-dosage group and the huperzine A group （P 〈 0.05）. The results of the Morris water maze and immunohistochemistry indicated that PNS can improve the capacity for spatial learning and memory in SAMP8 mice, and reduce the content of A β 1-40, A β 1-42 and expression of App in the brains of SAMP8 mice. The therapeutic effects of the PNS high-dosage group were greater than that of the PNS low-dosage group and the huperzine A group （P 〈 0.05）. CONCLUSION： These results support the hypothesis that PNS plays a therapeutic and protective role on the pathological lesions and learning dysfunction of Alzheimer＇s disease. The therapeutic effects of PNS for Alzheimer＇s disease are possibly achieved through downregulating the expression of the App gene and upregulating the expression of the Syp gene. The therapeutic effects of PNS are dose-dependent and are greater than the effect of huperzine A.

A call for enhanced data-driven insights into wind energy flow physics

With the increased availability of experimental measurements aiming at probing wind resources and wind turbine operations,machine learning(ML)models are poised to advance our understanding of the physics underpinning the interaction between the atmospheric boundary layer and wind turbine arrays,the generated wakes and their interactions,and wind energy harvesting.However,the majority of the existing ML models for predicting wind turbine wakes merely recreate Computational fluid dynamics(CFD)simulated data with analogous accuracy but reduced computational costs,thus providing surrogate models rather than enhanced data-enabled physics insights.Although ML-based surrogate models are useful to overcome current limitations associated with the high computational costs of CFD models,using ML to unveil processes from experimental data or enhance modeling capabilities is deemed a potential research direction to pursue.In this letter,we discuss recent achievements in the realm of ML modeling of wind turbine wakes and operations,along with new promising research strategies.

Parameter Self - Learning of Generalized Predictive Control Using BP Neural Network

This paper describes the self—adjustment of some tuning-knobs of the generalized predictive controller(GPC).A three feedforward neural network was utilized to on line learn two key tuning-knobs of GPC,and BP algorithm was used for the training of the linking-weights of the neural network.Hence it gets rid of the difficulty of choosing these tuning-knobs manually and provides easier condition for the wide applications of GPC on industrial plants.Simulation results illustrated the effectiveness of the method.

Imperatorin alleviates Aβ-induced spatial learning memory impairment and neuroinflam⁃mation in model mice of Alzheimer disease

OBJECTIVE To investigate the effects of imperatorin on the spatial learning memory impairment and neuroinflammation in model mice of Alzheimer disease(AD)induced by intracerebroventricular injection of Aβ1-42.METHODS Mouse model of AD was established by injection of Aβ1-42 into the lateral ventricles.Im⁃peratorin(2.5 and 5.0 mg·kg-1,daily)was inject⁃ed by intraperitoneally 1 h after intracerebroven⁃tricular injection for 13 d.The effect of imperato⁃rin on the spatial learning and memory impair⁃ment was assessed by eight arm maze tests.The levels of cytokines TNF-α,IL-1β,IL-6,IL-18 and chemokines MCP-1 in mouse cortex and hip⁃pocampus were detected by ELISA.The protein expression of NF-κB P65,TLR4,MyD88,p-P38,p-ERK,and p-JNK were detected by Western blotting.RESULTS As compared with the AD model group,imperatorin treatment significantly attenuated Aβ1-42-induced spatial learning and memory impairment assessed by eight arm maze tests.In addition,imperatorin significantly reduced the levels of cytokines TNF-α,IL-1β,IL-6,IL-18 and chemokines MCP-1 in the cerebral cortex and hippocampus.Meanwhile,Western blotting results showed that imperatorin treat⁃ment significantly down-regulated the protein expression of NF-κB P65,TLR4,MyD88,p-P38,p-ERK,and p-JNK.CONCLUSION Imperatorin has neuroprotective effects in the Aβ1-42 induced AD model mice and its mechanism may be partially associated with the inhibition of inflam⁃matory response in the cortex and hippocampus.

Adolescent Exposure of JWH-018 “Spice” Produces Subtle Effects on Learning and Memory Performance in Adulthood

The active components associated with the bio-designer drugs known variously as “Spice” or “K2” have rapidly gained in popularity among recreational users, forcing the United States Drug Enforcement Administration to classify these compounds as Schedule I drugs in the Spring of 2011. However, although there is some information about many of the synthetic cannabinoids used in Spice products, little is known about the consequences of the main constituent, (1-pentyl-3-(1-naphthoyl)indole;JWH-018), on neuropsychological development or behavior. In the present experiment, adolescent rats were given repeated injections of either saline or 100 μg/kg of JWH-018. Once the animals were 75 days of age, they were trained using tasks with spatial components of various levels of difficulty and a spatial learning set task. On early trials with water maze tasks of varying difficulty, the JWH-018 treated rats were impaired relative to controls. However, by the end of each phase of testing, drug and control animals were comparable, although on probe trials the drug-treated animals spent significantly less time in the target quadrant. In addition, the performance of the drug-treated rats was inferior to that of the control animals on a learning set task, suggesting some difficulty in adapting their responses to changing task demands. The results suggest that chronic exposure to this potent cannabinoid CB1 receptor agonist during adolescence is capable of producing a variety of subtle changes affecting spatial learning and memory performance in adulthood, well after the drug exposure period.

平台项目双引领及促进“科-产-教”深度融合

依托广州华立科技职业学院电力工业自动化研究院平台和广东省科技创新战略专项资金立项项目,创建平台、项目双引领。对接增城本地制造企业,提出构建Solid Learning(形塑学习)培养模式,该模式将数据库、企业、学校联系在一起,以解决企业难点、提升学生STEAM综合能力为目标,探索出一条具有区域特色的“科-产-教”深度融合发展之路。

NP-14 Effects of Osthole on the Improvement of Learning and Memory Impairment in A Mouse Model Injected with Aβ25-35

Objective:We aimed to investigate the effects of osthole on learning and memory impairment of AD mice induced by injection of Aβ25-35 and the content of Ca2+、GLU、Ab1-42 in the brain tissue and peripheral blood.Methods:Mice were randomly assigned to sham operation,Aβ25-35,Aβ25-35+Ost-L,Aβ25-35+Ost-M,and Aβ25-35+Ost-H group.Water maze test was performed to assessing spatial learning ability of mice.It is determined that the MDA level and the activity of SOD in the brain tissue of mice in each group by colorimetry.The GLU kit and Ca2+kit were used to detect the GLU,Ca2+in tissue and serum.Elisa was used to detect the expression of Aβ1-42 in the hippocampus and serum of mice.HE staining and silver staining were used to detect neuron apoptosis and pathological changes in brain slices.Results:①Effects of osthole on learning and memory:With the increase of training day,the escape latencies continuously reduced in each experimental group,the escape latencies of the model group was longer on the 1st,2nd,3rd,and 5th days than the normal group,the difference was statistically significant(day 3,4:P<0.05,day 5:P<0.01);compared with the model group,the escaping latency on the fifth day of the OST low-medium high-dose group was significantly shortened,which was statistically significant(P<0.05).②Effects on oxidative stresspathway:the SOD activity of AD mice in the hippocampus model group was lower than that in the normal group,which was statistically significant(P<0.05);The SOD activity in the OST group was higher than that in the model group,which was statistically significant(P<0.05).The MDA content in the model group was significantly higher than that in the normal group(P<0.05).The MDA content in the OST high-dose group was lower than that in the model group,which was statistically significant(P<0.05).③Effects of GLU levels on neurotransmitters:the results of the detection of GLU in cortical area and GLU in serum of AD mice in OST dose groups showed that serum GLU levels in the model group were significantly lower than those in the sham group,which was statistically significant(P<0.05).GLU levels in the cortical area were also significantly higher than those in the sham group,which was statistically significant(P<0.05).Compared with the model group,GLU levels in the OST administration group were significantly downregulated.Among the serum,the effect of medium dose group was obvious.Although there was a trend of down-regulation in the cortical administration group,there was no statistical significance.④Changes in Ca2+concentration in the brain;Detection of intracellular Ca ion concentration in AD mice by OST doses showed that,compared with the sham group,the model group was significantly upregulated in cortical Ca2+levels.There was no statistical difference in the administration group.Compared with the model group,the concentration of Ca2+in the OST-H group significantly decreased.⑤Effect on levels of Ab1-42 in hippocampus and serum:model group had significantly higher Ab1-42 levels in hippocampus than in sham operation group,which was statistically significant(P<0.05).Ab1-42 in serum was also significantly upregulated compared to the sham group,which was statistically significant(P<0.05).Compared with the model group,the levels of Aβ1-42 in the OST administration group were significantly down-regulated,with the lower and middle doses in the hippocampus being more significant,while the serum was more pronounced at lower doses.⑥Silver staining to detect the tangles of hippocampal neurons:Neuron tangles in the hippocampal CA1 region showed a dark brown-yellow granule distribution in the nuclei of the model group(positive expression).Nerve cell body and dendrites,axons are black or black red,background light yellow.Compared with the model group,the administration group has improved significantly.Conclusion:OST improves spatial learning and memory of dementia model mice injected with Ab25-35 in both hippocampus.Experimental studies have shown that OST has different degrees of regulation on neuronal apoptosis,Ca2+/GLU/oxidative stress and other pathways,and it plays a role in improving multiple AD pathological changes and delaying the pathogenesis of neurodegenerative diseases.

Design of Machine Learning Based Smart Irrigation System for Precision Agriculture

Agriculture 4.0,as the future of farming technology,comprises numerous key enabling technologies towards sustainable agriculture.The use of state-of-the-art technologies,such as the Internet of Things,transform traditional cultivation practices,like irrigation,to modern solutions of precision agriculture.To achieve effectivewater resource usage and automated irrigation in precision agriculture,recent technologies like machine learning(ML)can be employed.With this motivation,this paper design an IoT andML enabled smart irrigation system(IoTML-SIS)for precision agriculture.The proposed IoTML-SIS technique allows to sense the parameters of the farmland and make appropriate decisions for irrigation.The proposed IoTML-SIS model involves different IoT based sensors for soil moisture,humidity,temperature sensor,and light.Besides,the sensed data are transmitted to the cloud server for processing and decision making.Moreover,artificial algae algorithm(AAA)with least squares-support vector machine(LS-SVM)model is employed for the classification process to determine the need for irrigation.Furthermore,the AAA is applied to optimally tune the parameters involved in the LS-SVM model,and thereby the classification efficiency is significantly increased.The performance validation of the proposed IoTML-SIS technique ensured better performance over the compared methods with the maximum accuracy of 0.975.

基于GA优化GRU-LSTM-FC组合网络的风电场动态等值建模

针对风电场动态等值建模依赖于运行方式和特定扰动,难以获得普适性强的通用等值模型的难题,提出了基于门控循环单元-长短期记忆-全连接(GRU-LSTM-FC)组合网络的数据驱动建模方法,并提出基于遗传算法(GA)对组合网络模型进行调优。首先将风电机组描述为一组微分代数方程组,模型输入为测风塔风速、风向和公共耦合点处的电压时间序列,模型输出为风电场功率时间序列。然后对比了具有记忆作用的LSTM(GRU)网络结构与风电机组微分方程的相似性,以及FC网络结构与风电机组代数方程的相似性,提出基于GRU-LSTM-FC组合网络的风电场等值建模方法。为对组合网络进行模型调优,利用GA优化组合网络中的FC层数和各层神经元数目。最后以某风电场为例验证了所提组合网络进行风电场等值建模的可行性,并将所提方法与其他神经网络模型进行了对比,分析了所提模型的优越性。

Utilization Survey of Livestock Manure Resources in Large-scale Farms of Yangzhou

Based on surveying the conditions of large -scale farms and commercial manure in the each county of Yangzhou city, the situations and problems for utilization of livestock manure resources were grasped. After an analysis of the potential value of livestock manure, the suggestion and strategy for utilization of livestock manure resources were proposed based on the actual conditions in Yangzhou city.

Research on Reactive Power Optimization of Offshore Wind Farms Based on Improved Particle Swarm Optimization

The lack of reactive power in offshore wind farms will affect the voltage stability and power transmission quality of wind farms.To improve the voltage stability and reactive power economy of wind farms,the improved particle swarmoptimization is used to optimize the reactive power planning in wind farms.First,the power flow of offshore wind farms is modeled,analyzed and calculated.To improve the global search ability and local optimization ability of particle swarm optimization,the improved particle swarm optimization adopts the adaptive inertia weight and asynchronous learning factor.Taking the minimum active power loss of the offshore wind farms as the objective function,the installation location of the reactive power compensation device is compared according to the node voltage amplitude and the actual engineering needs.Finally,a reactive power optimizationmodel based on Static Var Compensator is established inMATLAB to consider the optimal compensation capacity,network loss,convergence speed and voltage amplitude enhancement effect of SVC.Comparing the compensation methods in several different locations,the compensation scheme with the best reactive power optimization effect is determined.Meanwhile,the optimization results of the standard particle swarm optimization and the improved particle swarm optimization are compared to verify the superiority of the proposed improved algorithm.

Expression of miR-9-5p and RHOA in aluminum-induced rat cognitive dysfunction

Objective:To investigate the possible mechanism of microRNA-9-5p(miR-9-5p)and Ras homologous gene family A(RHOA)in aluminum-induced cognitive dysfunction in rats.Methods:According to the principle of randomization,48 Wistar rats were randomly divided into four groups(n=12)of blank control,low dose,medium dose and high dose.The blank control group was gavaged daily saline,and the other three dose groups were given daily gavage AlCl3 aqueous solution at three doses of 25 mg/kg,50 mg/kg,and 100 mg/kg to create a rat model of cognitive impairment for three months.The water maze(MWM)positioning navigation experiment was used to record the time t(s),namely,the incubation period,on the platform of rats,and the incubation period of each group was used to determine whether the rats in the infected group had learning and memory impairment.Hematoxylin-eosin(HE)and Nissl stains observed the pathological changes of nerve cells in the hippocampus of the four groups.Western blot detected the protein expression levels of RHOA and cranial neurotrophic factor(BDNF)in fresh rat hippocampal tissues.RT-qPCR detected the mRNA expression of miR-9-5p,RHOA,and BDNF in rat hippocampal tissues.Results:The results of Morris water maze positioning navigation test showed that the incubation period of each group was calculated on the 1st,3rd and 5th days of the experiment,and the motor incubation period of the infected group was higher than that of the control group.The results of HE staining showed that the rat nerve cells in the control group were morphologically intact,the staining was clear,the nucleus was clearly visible,and the edge of the cell membrane was sharp.The rat neurons in the infected group were damaged to varying degrees,the nucleus gradually dissolved,the cytoplasmic staining became deeper,the edges of the cell membrane were blurred and disordered,and the cells were deformed and arranged disordered.The results of Nissl staining showed that the well-stained Nissl body particles were visible in the nerve cells of rats in the control group,and the dissipation of Nissl bodies in the nerve cells of the infected group was reduced,and the staining was shallow.The results of RT-qPCR showed that compared with the control group,the mRNA expression of miR-9-5p and BDNF was decreased in the infected group,and the mRNA expression of RHOA was increased(P<0.05 or P<0.001).The Western blot results showed that compared with the control group,the relative expression of BDNF in the three infected groups was decreased,and the relative expression of RHOA increased(P<0.05).Conclusion:In aluminum-induced cognitive impairment,miR-9-5p is downregulated and RHOA is upregulatd.

基于深度学习的鱼类养殖监测研究进展

鱼类养殖是通过人工方式在水中养殖各种鱼类的经济活动。鱼类养殖可以在淡水、海水或者盐碱水环境中进行,通过各种监测技术和设备来培育和管理鱼的生长和繁殖。传统的鱼类养殖监测方法存在效率低和准确性差等问题。近年来,基于深度学习的视觉技术的发展为鱼类养殖监测提供了新的解决方案。该文阐述了基于深度学习的视觉技术在鱼类养殖监测中的应用,并从鱼体测量、鱼类计数、鱼类摄食、鱼类游泳行为和鱼病诊断5个方面分别对研究进展进行梳理。在此基础上总结了鱼类养殖监测在数据采集与传输、建立鱼类养殖监测数据集、超规模参数模型、终端监测设备边缘计算、数字孪生、智能监测业务化应用不足等问题和展望,旨在为深度学习在鱼类养殖监测中的推广应用提供科学参考。

基于残差神经网络的鸡蛋分类识别研究

【目的】探究残差神经网络(residual neural network,ResNet)对不同种类鸡蛋的分类效果,明确深度学习应用存在智能鸡蛋巡检装置的可行性,为家禽养殖智能化进程提供新思路,并为鸡蛋分类研究提供数据支撑。【方法】在鸡舍实地取样,采用自适应矩估计优化器(adaptive moment estimation,Adam)以微调最后1层、微调所有层和重新训练所有层3种迁移学习策略分别训练,并通过调整模型权重参数及改变学习率的方式训练出最佳分类模型。【结果】得到识别准确率高达98.971%的鸡蛋分类模型。计算出模型在数据集上的各类评估指标,并借助混淆矩阵及语义特征降维可视化,分析出鸡蛋分类识别中易被误判的类别及语义。该模型部署后实时性良好,满足实际需求。【结论】鸡蛋的分类识别中光照条件是关键影响因素,应尽可能使鸡舍光照稳定均衡。针对6类鸡蛋,微调所有层并调整学习率参数为0.6,可得最佳模型。其在鸡舍场景下分类效果优良,尤其是颜色语义,应用于智能鸡蛋巡检装置,可有效降低人力成本。后续研究中应注重畸形蛋及软壳蛋的记录,为进一步优化提供数据支撑。