基于注意力机制的林木物候期识别方法

【目的】针对物候期识别传统方法特征提取不充分、未对关键特征进行区分,导致方法泛化能力较差、迁移应用识别精度低的问题,本研究将注意力机制引入残差神经网络,结合基于数字照相的物候观测方式,提出具有较强细粒度特征识别能力且实用性较强的林木物候期识别方法,从而为林木的长期实时物候监测提供技术支撑。【方法】以PhenoCam中的栎林及槭林像片为研究材料,选取2017—2018年的数据作为训练集,以2019年的数据评价模型的泛化能力。研究结合实地观测数据对研究区的林木物候期进行划分,设计数据裁剪公式,在增强数据的同时均衡各个物候期数据的数量。研究基于ResNet50残差神经网络设计了深度学习模型,针对林木物候期差异较小的特性引入了注意力机制,注意力模块在通道及空间维度对神经网络提取的特征进行再处理,提升了模型对细粒度图像差异的识别能力。【结果】注意力机制的引入有效提升了模型的泛化能力,增强了模型对易混淆物候期的识别能力,在未参与训练的19年数据集的栎林物候期识别取得了90.58%的准确率,槭林物候期识别准确率为89.27%,较引入前模型准确率在两个研究区分别提升21.86%与13.15%,优于传统AlexNet和HOG-SVM方法。【结论】基于残差注意力网络的林木物候期识别方法较好解决了原有方法泛化能力较差的问题,该方法具有准确度高、迁移应用能力强等优势,能对易混淆的林木物候期进行精准区分,适用于林木物候的长期观测。

基于改进YOLO v8-Pose的红熟期草莓识别和果柄检测

针对高架栽培模式下的大棚草莓,借鉴人体姿态检测算法,建立了改进YOLO v8-Pose模型对红熟期草莓进行识别与果柄关键点检测。通过对比YOLO v5-Pose、YOLO v7-Pose、YOLO v8-Pose模型,确定使用YOLO v8-Pose模型作为对红熟期草莓识别与关键点预测的模型。以YOLO v8-Pose为基础,对其网络结构添加Slim-neck模块与CBAM注意力机制模块,提高模型对小目标物体的特征提取能力,以适应草莓数据集的特点。改进YOLO v8-Pose能够有效检测红熟期草莓并准确标记出果柄关键点,P、R、mAP-kp分别为98.14%、94.54%、97.91%,比YOLO v8-Pose分别提高5.41、5.31、8.29个百分点。模型内存占用量为22 MB,比YOLO v8-Pose的占用量小6 MB。此外,针对果园非结构化的特征,探究了光线、遮挡与拍摄角度对模型预测的影响。对比改进前后的模型在复杂环境下对红熟期草莓的识别与果柄预测情况,改进YOLO v8-Pose在受遮挡、光线和角度影响情况下的mAPkp分别为94.52%、95.48%、94.63%,较YOLO v8-Pose分别提高8.9、10.75、5.17个百分点。改进YOLO v8-Pose可在保证网络模型精度的同时对遮挡、光线和拍摄角度等影响均具有较好的鲁棒性,能够实现对复杂环境下红熟期草莓识别与果柄关键点预测。

基于时序PolSAR影像与决策树模型的油菜物候期识别

作物物候期识别是农情遥感监测的重要内容,及时准确识别作物物候期,对有效评估作物生长趋势、提高农情信息化管理水平有重要意义。提出了基于时间序列全极化合成孔径雷达(polarimetric synthetic aperture radar,PolSAR)数据结合决策树模型的油菜物候期识别方法。首先,采用3种极化分解方法提取PolSAR极化参数,并分析各极化参数对油菜物候期的动态响应规律;其次,基于各极化分解方法提取的参数建立决策树模型,并对油菜物候期进行分类识别;最后,采用基于混淆矩阵的方法对油菜物候期识别结果进行精度评价。采用5期Radarsat-2 PolSAR数据和地面物候观测数据进行实验验证。结果表明:提取的PolSAR参数中对物候期变化较为敏感的参数有H/A/alpha分解中的散射角(Alpha)、特征值(L2、L3)、伪熵(P2)、目标方位角(Beta1)参数,Freeman-Durden分解中的地面散射(Ground)和奇次散射(Odd)参数,Yamaguchi分解中的奇次散射(Odd_Y)和螺旋体散射(Helix)参数;决策树模型对油菜物候期识别结果较为准确,识别结果中组合3种极化分解方法提取参数建立的原始决策树模型分类总体精度最高,达94%。总体上,PolSAR极化分解参数对油菜物候期变化比较敏感,决策树模型能有效识别油菜物候期。

基于改进ResNet50模型的自然环境下苹果物候期识别

[目的/意义]针对传统方法对自然环境下苹果物候期图像识别精度低、覆盖面不全等问题,提出一种基于改进ResNet50模型的苹果物候期识别方法。[方法]通过搭建球形摄像机获取复杂背景下的苹果图像数据集,以ResNet50作为基础模型,引入SE (Squeeze-and-Excitation Network)通道注意力机制强化对苹果图像的特征提取能力,并结合余弦退火衰减学习率的Adam优化器,实现自然环境下高原红富士苹果物候期图像的智能识别。[结果]在32,000幅苹果树图像集上进行了试验,结果表明,改进ResNet50模型对苹果物候期图像进行识别,验证集准确率达到96.35%,测试集准确率达到91.94%,平均检测时间为2.19 ms,相较于AlexNet、VGG16、ResNet18、ResNet34、ResNet101以及经典ResNet50模型,最优验证集准确率分别提升了9.63%、5.07%、5.81%、4.55%、0.96%和2.33%。[结论]改进ResNet50可实现对苹果物候期有效识别,该研究成果可为果园物候期识别提供参考,通过集成至果树生育期智能监测生产管理平台,实现苹果园区的智能化管控。

改进的轻量级YOLO在苹果物候期自动观测中的研究

为在复杂的田间场景下实时检测果树上苹果并识别其所处的物候期(幼果期、果实膨大期、果实成熟期),提出一种用于嵌入式物候期监测设备的改进的Tiny-YOLOv3模型。针对多样复杂的物候期图像,提出适度深化Tiny-YOLOv3模型,在深化后的模型上应用DenseNet方法强化低分辨率层特征提取,在实时条件下提高各个物候期检测识别的平均精度。实验结果表明,与Tiny-YOLOv3相比,所提Minor-YOLOv3-dense模型的检测识别精度提高了5.56%,检测速度略有下降,对整个模型的实时性没有特别的影响。

白洋淀湿地生态水位及生态补水方案分析

入淀水量不足,是造成白洋淀湿地生态及其环境恶化,淀泊萎缩的重要原因,如何修复已遭破坏的白洋淀湿地生态系统,并维护湿地的生态平衡是亟待解决的重要问题。本研究采用白洋淀湿地从1956—2000年的逐月实测水位数据,通过分析汛期和非汛期的水位经验频率以及白洋淀湿地的生态功能,确定了白洋淀周年的生态适宜水位范围为7.9-8.7 m。当白洋淀出现生态干旱,水位不能维持在7.9-8.7 m时,生态补水后应使淀水位保持在7.1-7.9 m范围内。本文从补水量和补水时间2个方面,对白洋淀湿地的生态补水方案进行了研究。

广聚萤叶甲成虫取食经历对后续产卵寄主选择的影响

【背景】取食经历对植食性昆虫的寄主选择行为具有较大影响,影响天敌昆虫寄主专一性测定的设计和结果解释。【方法】采用选择性试验,观察了入侵豚草的重要天敌——广聚萤叶甲成虫羽化后取食不同植物对其后续产卵寄主选择的影响。【结果】与取食豚草的试虫相比,有取食三裂叶豚草、苍耳或菊芋经历的成虫选择苍耳产卵的频次增加,不再对豚草表现出明显的选择偏好性。对产卵识别期的Cox模型分析结果表明,成虫早期取食不同植物,对后续产卵选择有显著影响,成虫羽化后如果先取食豚草或三裂叶豚草,则选择苍耳产卵的倾向显著低于豚草;但如果先取食苍耳、菊芋和农家向日葵,则选择苍耳产卵的倾向与豚草无显著差异。【结论与意义】由此推测,广聚萤叶甲初羽化成虫取食的植物对其后续产卵选择具有较大影响,因而在寄主专一性测定中应关注测试前饲喂的植物种类。

基于特征优选决策树模型的河套灌区土地利用分类

为了提高土地利用遥感识别精度,探索不同识别期及不同特征变量对土地利用类型遥感识别精度的影响。该研究采用Landsat时间序列影像数据,考虑不同月份和不同特征变量(波段、光谱指数及纹理特征)组合方式建立土地利用决策树分类模型,并利用河套灌区永济灌域实测数据和Google earth影像对不同组合方式的土地利用模型进行数量结构和空间布局的验证,筛选出最优的土地利用遥感模型并确定最佳识别期。结果表明:在不同月份Green(绿波段)和Ent(熵Entropy)分别与波段和纹理特征变量中的因子所含有的信息重复率最高,需剔除,归一化植被指数(Normalized Differential Vegetation Index,NDVI)和增强型植被指数(Enhanced Vegetation Index,EVI)在今后的研究中可选其一应用;与单一特征变量相比,不同特征变量组合后能提高模型精度,平均总体精度和Kappa系数分别提高了6.72个百分点和0.09。采用8月影像数据构建的遥感模型精度最高,最优遥感模型的特征变量组合方式为波段+光谱指数+纹理特征,总体精度、Kappa系数、制图精度和用户精度分别为80.23%、0.74、80.95%和86.26%,且减少了未利用地和居民工况用地空间布局的错分。通过综合比较,该研究区土地利用最佳识别期为8月,其次为9月。利用8月最优遥感模型(最佳识别期和最优组合)识别的耕地、林地、草地、未利用地、水域和居民工矿用地的制图精度分别为96.83%、73.33%、70.00%、65.52%、100.00%和80.00%,用户精度分别为76.62%、100.00%、82.35%、82.61%、100.00%和80.00%。因此可选用8月最优模型应用于长时间序列的土地利用类型识别。

Application of Hilbert-Huang transform based delay time identification in optimization of short millisecond blasting

The accurate identification of delay time in millisecond blasting plays an important role in the optimization of blasting design and reduction of vibration effect. Through a case study of a surge shaft blasting project, the capability of the EMD （empirical mode decomposition） method in identifying the delay time of short millisecond blasting with precise initiation was compared with the instantaneous energy method based on Hilbert-Huang transform （HHT）. The recognition rate of the EMD method was more than 80%, while the instantaneous energy method was less than 25%. By analyzing the instantaneous energy of single-hole blasting signal, it was found that the instantaneous energy method was adaptable to millisecond blasting with delay time longer than half of the energy peak period. The EMD method was used to identify delay time of millisecond blasting in Zijinshan open-pit mine. According to the identification results, the blasting parameters were optimized for controlling the blast-induced vibration and reducing the large block ratio. The field data showed that the velocity peak of ground vibration was reduced by more than 30%under almost the same maximum charge per delay by the optimization of delay time and detonating detonators. Combining with slag-remaining blasting and burden optimization of the first row, the large block ratio was reduced to less than 3%. The research results proved that the identification method based on HHT was feasible to optimize the blasting design. The identification method is also of certain reference value for design optimization of other similar blasting projects.

Current trends in staging rectal cancer

Management of rectal cancer has evolved over the years. In this condition preoperative investigations assist in deciding the optimal treatment. The relation of the tumor edge to the circumferential margin (CRM) is an important factor in deciding the need for neoadjuvant treatment and determines the prognosis. Those with threatened or involved margins are offered long course chemoradiation to enable R0 surgical resection. Endoanal ultrasound (EUS) is useful for tumor (T) staging; hence EUS is a useful imaging modality for early rectal cancer. Magnetic resonance imaging (MRI) is useful for assessing the mesorectum and the mesorectal fascia which has useful prognostic significance and for early identification of local recurrence. Computerized tomography (CT) of the chest, abdomen and pelvis is used to rule out distant metastasis. Identification of the malignant nodes using EUS, CT and MRI is based on the size, morphology and internal characteristics but has drawbacks. Most of the common imaging techniques are suboptimal for imaging following chemoradiation as they struggle to differentiate fibrotic changes and tumor. In this situation, EUS and MRI may provide complementary information to decide further treatment. Functional imaging using positron emission tomography (PET) is useful, particularly PET/CT fusion scans to identify areas of the functionally hot spots. In the current state, imaging has enabled the multidisciplinary team of surgeons, oncologists, radiologists and pathologists to decide on the patient centered management of rectal cancer. In future, functional imaging may play an active role in identifying patients with lymph node metastasis and those with residual and recurrent disease following neoadjuvant chemoradiotherapy.

Update on risk scoring systems for patients with upper gastrointestinal haemorrhage

Upper gastrointestinal haemorrhage (UGIH) remains a common medical emergency worldwide. It is increasingly recognised that early risk assessment is an important part of management, which helps direct appropriate patient care and the timing of endoscopy. Several risk scores have been developed, most of which include endoscopic findings, although a minority do not. These scores were developed to identify various end-points including mortality, rebleeding or clinical intervention in the form of transfusion, endoscopic therapy or surgery. Recent studies have reported accurate identification of a very low risk group on presentation, using scores which require simple clinical or laboratory parameters only. This group may not require admission, but could be managed with early out-patient endoscopy. This article aims to describe the existing pre- and post-endoscopy risk scores for UGIH and assess the published data comparing them in the prediction of outcome. Recent data assessing their use in clinical practice, in particular the early identification of low-risk patients, are also discussed.

利用典型Stokes参数的油菜物候期识别

油菜关键物候期信息的获取对于油菜的田间管理、观赏时间预测及产量估测等具有重要意义,是精准农业实施的重要组成部分。极化合成孔径雷达技术不仅可以实现对作物全天时监测,而且对作物的结构信息敏感,在物候期提取中极具潜力。首先,以覆盖油菜整个生长期的5景时间序列全极化Radarsat-2数据为基础,基于Stokes矢量提取了平均强度g0、归一化平均强度g0m、平均极化度ρm、零度方向路线球面度Pdor、零度孔径路线倾斜度Idap和零度孔径路线弧对称度Aadap6个典型Stokes参数;然后,对比分析了这6个参数对油菜整个生长期动态变化的响应特征,并以此为基础采用决策树(decision tree,DT)算法对油菜的物候期进行了识别。研究结果表明,6个Stokes参数中,除ρm和Aadap外,其他4个参数均对油菜物候期变化敏感,在油菜物候期识别中具有极大的潜力。DT算法能有效识别油菜的各关键物候期,其分类结果与样地实测数据具有良好的一致性,总体分类精度为87.4%;在单个物候期的识别中,识别精度最高达到了94.3%。

Multi-head attention-based long short-term memory model for speech emotion recognition

To fully make use of information from different representation subspaces,a multi-head attention-based long short-term memory(LSTM)model is proposed in this study for speech emotion recognition(SER).The proposed model uses frame-level features and takes the temporal information of emotion speech as the input of the LSTM layer.Here,a multi-head time-dimension attention(MHTA)layer was employed to linearly project the output of the LSTM layer into different subspaces for the reduced-dimension context vectors.To provide relative vital information from other dimensions,the output of MHTA,the output of feature-dimension attention,and the last time-step output of LSTM were utilized to form multiple context vectors as the input of the fully connected layer.To improve the performance of multiple vectors,feature-dimension attention was employed for the all-time output of the first LSTM layer.The proposed model was evaluated on the eNTERFACE and GEMEP corpora,respectively.The results indicate that the proposed model outperforms LSTM by 14.6%and 10.5%for eNTERFACE and GEMEP,respectively,proving the effectiveness of the proposed model in SER tasks.

Jejunoileal bypass:A surgery of the past and a review of its complications

Jejunoileal bypass(JIB),popular in the 1960s and 1970s,had remarkable success in achieving weight loss by creating a surgical short bowel syndrome.Our patient had an unusual case of liver disease and provided no history of prior bariatric surgery.Later,it was recognized that he had a JIB in the 1970s,which was also responsible for the gamut of his illnesses.Patients with JIB are often not recognized,as they died of complications,or underwent reversal of their surgery or a liver-kidney transplant.Early identification with prompt reversal,and the recognition and treatment of the life-threatening consequences play a critical role in the management of such patients.

Magnets, children and the bowel: A dangerous attraction?

Reports of magnet ingestion are increasing rapidly globally. However, multiple magnet ingestion, the subsequent potential complications and the importance of the early identification and proper management remain both under-recognized and underestimated. Published literature on such cases could possibly represent only the tip of an iceberg with press reports, web blogs and government documents highlighting further occurrence of many more such incidents. The increasing number of complications worldwide being reported secondary to magnet ingestion point not only to an acute lack of awareness about this condition among the medical profession but also among parents and carers who will be in most cases the first to pick up on magnet ingestion. There still seems to be no consensus on the management of magnet ingestion with several algorithms being proposed for management. Prevention of this condition remains a much better option than cure. Proper education and improved awareness among parents and carers and frontline medical staff is key in addressing this rapidly emerging problem. The goal of managing such cases of suspected magnet ingestion should be aimed at reducing delays between ingestion time, diagnosis time and intervention time.

How Many Packets Are Most Effective for Early Stage Traffic Identification： An Experimental Study

Accurately identifying network traffics at the early stage is very important for the application of traffic identification.Recent years,more and more research works have tried to build effective machine learning models to identify traffics with the few packets at the early stage.However,a basic and important problem is still unresolved,that is how many packets are most effective in early stage traffic identification.In this paper,we try to resolve this problem using experimental methods.We firstly extract the packet size of the first 2-10 packets of 3 traffic data sets.And then execute crossover identification experiments with different numbers of packets using 11 well-known machine learning classifiers.Finally,statistical tests are applied to find out which number is the best performed one.Our experimental results show that 5-7are the best packet numbers for early stage traffic identification.

Genetic-fuzzy HEV control strategy based on driving cycle recognition

A genetic-fuzzy HEV control strategy based on driving cycle recognition （DCR） was built. Six driving cycles were selected to represent different traffic conditions e.g. freeway, urban, suburb. A neural algorithm was used for traffic condition recognition based on ten parameters of each driving cycle. The DCR was utilized for optimization of the HEV control parameters using a genetic-fuzzy approach. A fuzzy logic controller （FLC） was designed to be intelligent to manage the engine to work in the vicinity of its optimal condition. The fuzzy membership function parameters were optimized using the genetic algorithm （GA） for each driving cycle. The result is that the DCR_ fuzzy controller can reduce the fuel consumption by 1. 9%, higher than only CYC _ HWFET optimized fuzzy （0.2%） or CYC _ WVUSUB optimized fuzzy （0.7%）. The DCR_ fuzzy method can get the better result than only optimizing one cycle on the complex real traffic conditions.

Image Segmentation Based on Period Difference of the Oscillation

A new method for image segmentation based on pulse neural network is proposed. Every neuron in the network represents one pixel in the image and the network is locally connected. Each group of the neurons that correspond to each object synchronizes while different groups of the neurons oscillate at different period. Applying this period difference, different objects are divided. In addition to simulation, an analysis of the mechanism of the method is presented in this paper.

An early recognition algorithm for BitTorrent traffic based on improved K-means

In response to the deficiencies of BitTorrent, the concept of density radius was proposed, and the distance from the maximum point of radius density to cluster center as a cluster radius was taken to solve the too large cluster radius resulted from the discrete points and to reduce the false positive rate of early recognition algorithms. Simulation results show that in the actual network environment, the improved algorithm, compared with K-means, will reduce the false positive rate of early identification algorithm from 6.3% to 0.9% and has a higher operational efficiency.