Comparison of artificial neural network and linear regression models for prediction of ring spun yarn properties. I. Prediction of yarn tensile properties

In this study artificial neural network (ANN) models have been designed to predict the ring cotton yarn properties from the fiber properties measured on HVI (high volume instrument) system and the performance of ANN models have been compared with our previous statistical models based on regression analysis. Yarn count, twist and roving properties were selected as input variables as they give significant influence on yarn properties. In experimental part, a total of 180 cotton ring spun yarns were produced using 15 different blends. The four yarn counts and three twist multipliers were chosen within the range of Ne 20–35 and α _e 3.8–4.6 respectively. After measuring yarn tenacity and breaking elongation, evaluations of data were performed by using ANN. Afterwards, sensitivity analysis results and coefficient of multiple determination (R²) values of ANN and regression models were compared. Our results show that ANN is more powerful tool than the regression models.     

Improving the accuracy of models to map alpine grassland above-ground biomass using Google earth engine

Accurate modelling and mapping of alpine grassland aboveground biomass (AGB) are crucial for pastoral agriculture planning and management on the Qinghai Tibet Plateau (QTP). This study assessed the effectiveness of four popular models (traditional multiple linear regression (MLR), support vector machine (SVM), artificial neural network (ANN), and deep neural network (DNN)) with various input combinations (geospatial variables [GV], vegetation types [VT], field measurements [FM], meteorological variables [MV] and observation time [OT]) for AGB estimation based on a new framework for AGB modelling and mapping using Google Earth Engine. The results showed that the input feature of GV had a poor performance in AGB estimation (0.121 < R² < 0.591). FM improved the accuracy the most when incorporated with GV (0.815 < R² < 0.833). Although MV, VT and OT improved the accuracy (R²) only by 0.112–0.216 with an importance rank order of MV > VT > OT for machine learning models, their outputs could be used to map AGB. Grass AGB was less accurately predicted than shrub AGB, but the pooling of both VTs improved estimation accuracy (R²) by 0.171–0.269. The performance of the models followed the ranked order of DNN > ANN > SVM > MLR. DNN had the highest accuracy (R² = 0.818) using all non-field measured variables (excluding FM) as the inputs, and it was successfully applied to a new dataset (not associated with the data used in the training and testing) with a R² of 0.676. This study presents an effective and operational framework for modelling and mapping grassland AGB. Accordingly, it provides the scientific foundations to determine of sustainable grazing carrying capacity in alpine grasslands.     

Predicting properties of single jersey fabrics using regression and artificial neural network models

In our previous works, we had predicted cotton ring yarn properties from the fiber properties successfully by regression and ANN models. In this study both regression and artificial neural network has been applied for the prediction of the bursting strength and air permeability of single jersey knitted fabrics. Fiber properties measured by HVI instrument and yarn properties were selected as independent variables together with wales’ and courses’ number per square centimeter. Firstly conventional ring yarns were produced from six different types of cotton in four different yarn counts (Ne 20, Ne 25, Ne 30, and Ne 35) and three different twist multipliers (α _e 3.8, α _e 4.2, and α _e 4.6). All the yarns were knitted by laboratory circular knitting machine. Regression and ANN models were developed to predict the fabric properties. It was found that all models can be used to predict the single jersey fabric properties successfully. However, ANN models exhibit higher predictive power than the regression models.     

Prediction of the changes on the CIELab values of fabric after chemical finishing using artificial neural network and linear regression models

Changes on the CIELab values of the dyed materials after the different chemical finishing treatments using artificial neural network (ANN) and linear regression (LR) models have been predicted. The whole structural properties of fabrics and some process data which were from fiber to the finishing parameters were accepted as inputs in these models. The networks having different structures were established, and it was also focus on the parameters which could affect the performance of the established networks. It was determined that we could successfully predict the color differences values occurring on the material after the finishing applications. In addition, we realized that some ANN parameters affected the prediction performance while establishing the models. After training ANN models, the prediction of the color difference values was also tried by linear regression models. Then, extra ANN models were established for all outputs using the parameters as inputs in the LR equations, and the prediction performances of both established models were compared. According to the results, the neural network model gives a more accurate prediction performance than the LR models.     

Modelling of ring yarn unevenness by soft computing approach

This paper demonstrates the application of two soft computing approaches namely artificial neural network (ANN) and neural-fuzzy system to forecast the unevenness of ring spun yarns. The cotton fiber properties measured by advanced fiber information system (AFIS) and yarn count have been used as inputs. The prediction accuracy of the ANN and neural-fuzzy models was compared with that of linear regression model. It was found that the prediction performance was very good for all the three models although ANN and neural-fuzzy models seem to have some edge over the linear regression model. The linguistic rules developed by the neural-fuzzy system unearth the role of input variables on the yarn unevenness.     

Using artificial neural network to predict colour properties of laser-treated 100% cotton fabric

In this paper, artificial neural network (ANN) model was used for predicting colour properties of 100 % cotton fabrics, including colour yield (in terms of K/S value) and CIE L, a, and b values, under the influence of laser engraving process with various combination of laser processing parameters. Variables examined in the ANN model included fibre composition, fabric density (warp and weft direction), mass of fabric, fabric thickness and linear density of yarn (warp and weft direction). The ANN model was compared with a linear regression model where the ANN model produced superior results in prediction of colour properties of laser engraved 100 % cotton fabrics. The relative importance of the examined factors influencing colour properties was also investigated. The analysis revealed that laser processing parameters played an important role in affecting the colour properties of the treated 100 % cotton fabrics.     

Prediction of CIELab data and wash fastness of nylon 6,6 using artificial neural network and linear regression model

We tried to predict the CIELab data and wash fastness values of scoured nylon 6.6 knitted fabric dyed with 1:2 metal-complex acid dyes and aftertreated using three different methods named as syntan, syntan/cation and full backtan by artificial neural network (ANN) with Levenberg-Marquardt algorithm and regression models. Afterward, the predicting performance of these models was tested and compared with each other using unseen data sets. We were able to achieve to predict the all colorimetric data satisfactorily such as L*, a*, b*, C, h ^o and wash fastness performance using both models. The statistical findings indicated that the regression models provide more accurate prediction for all colour data with an average error of 1% contrast to previous study. In terms of prediction of fastness, artificial neural network is a bit more useful than regression models for prediction of staining value on the nylon part of adjacent multifiber.     

The use of hybrid algorithms to improve the performance of yarn parameters prediction models

Although gradient based Backpropagation (BP) training algorithms have been widely used in Artificial Neural Networks (ANN) models for the prediction of yarn quality properties, they still suffer from some drawbacks which include tendency to converge to local minima. One strategy of improving ANN models trained using gradient based BP algorithms is the use of hybrid training algorithms made of global based algorithms and local based BP algorithms. The aim of this paper was to improve the performance of Levenberg-Marquardt Backpropagation (LMBP) training algorithm, which is a local based BP algorithm by using a hybrid algorithm. The hybrid algorithms combined Differential Evolution (DE) and LMBP algorithms. The yarn quality prediction models trained using the hybrid algorithms performed better and exhibited better generalization when compared to the models trained using the LM algorithms.     

基于等距拆分和随机森林算法的皖北小麦始花期气象预报

为探讨基于等距拆分和随机森林算法用于皖北小麦始花期气象预报的可行性,利用1980-2019年皖北地区7个农业气象观测站的冬小麦始花期原位观测物候数据和平行观测的气象数据,采用相关系数法,筛选影响始花期早迟的特征变量,采用有序等距离抽样法,拆分出训练集和测试集。基于随机森林算法（RF）,从4月10日到4月15日,每日训练1个预报模型,实现小麦始花期逐日滚动气象预报,并与基于类神经网络（ANN）、线性支撑向量机（LSVM）、多元回归（RG）和支持向量机（SVM）4种算法训练的预报模型进行比较。结果表明,由平均气温、最高气温、日照时数3类气象要素构成的40个关键气象因子与小麦始花期早迟密切相关;训练出的6个始花期逐日气象预报模型中,4月10-14日5个模型入选特征变量均为40个,4月15日模型入选特征变量为39个;6个气象预报模型训练集与测试集的平均正确率分别为93.3%和80.4%,平均均方根误差（RMSE）分别为1.860～1.960和2.510～2.709,平均决定系数分别为0.944和0.841;基于RF算法训练的预报模型3项检验指标均优于ANN、LSVM、RG和SVM算法训练的预报模型;利用RF算法模型在2020年和2021年进行预报,提前7～9 d准确预报出当年始花期。由此可见,采用有序等距离抽样拆分出训练集,再基于RF算法构建的皖北地区小麦花期气象预报模型,能够以较高精度对小麦始花期进行预报。     

Prediction of rotor spun yarn strength using support vector machines method

A new method for rotor spun yarn prediction from fiber properties based on the theory of support vector machines (SVM) was introduced. The SVM represents a new approach to supervised pattern classification and has been successfully applied to a wide range of pattern recognition problems. In this study, high volume instrument (HVI) and advanced fiber information system (Uster AFIS) fiber test results consisting of different fiber properties are used to predict the rotor spun yarn strength. The results obtained through this study indicated that the SVM method would become a powerful tool for predicting rotor spun yarn strength. The relative importance of each fiber property on the rotor spun yarn strength is also expected. The study shows also that the combination of SVM parameters and optimal search method chosen in the model development played an important role in better performance of the model. The predictive performances are estimated and compared to those provided by ANFIS model.     

Analysis and tensile-tear properties of abraded denim fabrics depending on pattern relations using statistical and artificial neural network models

The aim of this study is to develop new pattern denim fabrics and characterize the mechanical properties of these fabrics after abrasion load. Furthermore, tensile and tear strengths of these fabrics have been analysed by using the Artificial Neural Network (ANN) and statistical model. All denim fabrics were first abraded and subsequently tensile and tearing tests were applied to the abraided fabrics seperately. Actual data generated from the tests were analyzed by ANN and regression model. The regression model has shown that tensile strength properties of the abraded large structural pattern denim fabrics are generally low compared to that of the small structural pattern and traditional denim fabrics. On the other hand, when the abrasion cycles are increased tensile properties of all denim fabrics are generally decreased. Tearing strength of weft and warp in the abraded large structural pattern denim fabrics are between small structural pattern and traditional denim fabric. On the other hand, when the abrasion cycles are increased tearing strength properties in the weft and warp for all denim fabrics are generally decreased. The results from ANN and regression models were also compared with the measured values. It is concluded that almost all values from ANN are accurately predicted compared with those of the regression model. Therefore, we suggest that both methods can be used in this study as viable and reliable tools.     

Application of machine-learning models for diagnosing health hazard of nitrate toxicity in shallow aquifers

There is a growing concern about health hazards linked to nitrate (NO₃) toxicity in groundwater due to overuse of nitrogen fertilizers in rice production systems of northern Iran. Simple-cost-effective methods for quick and reliable prediction of NO₃ contamination in groundwater of such agricultural systems can ensure sustainable rural development. Using 10-year time series data, the capability of adaptive neuro-fuzzy inference system (ANFIS) and support vector machine (SVM) models as well as six geostatistical models was assessed for predicting NO₃ concentration in groundwater and its noncarcinogenic health risk. The dataset comprised 9360 water samples representing 26 different wells monitored for 10 years. The best predictions were found by SVM models which decreased prediction errors by 42–73 % compared with other models. However, using well locations and sampling date as input parameters led to the best performance of SVM model for predicting NO₃ with RMSE = 4.75–8.19 mg l^?1 and MBE = 3.3–5.2 mg l^?1. ANFIS models ranked next with RMSE = 8.19–25.1 mg l^?1 and MBE = 5.2–13.2 mg l^?1 while geostatistical models led to the worst results. The created raster maps with SVM models showed that NO₃ concentration in 38–97 % of the study area usually exceeded the human-affected limit of 13 mg l^?1 during different seasons. Generally, risk probability went beyond 90 % except for winter when groundwater quality was safe from nitrate viewpoint. Noncarcinogenic risk exceeded the unity in about 1.13 and 6.82 % of the study area in spring and summer, respectively, indicating that long-term use of groundwater poses a significant health risk to local resident. Based on the results, SVM models were suitable tools to identify nitrate-polluted regions in the study area. Also, paddy fields were the principal source of nitrate contamination of groundwater mainly due to unmanaged agricultural activities emphasizing the importance of proper management of paddy fields since a considerable land in the world is devoted to rice cultivation.     

Modeling spun yarns migratory properties using artificial neural network

The mechanical and physical properties of spun yarns and fabrics depend not only on properties of constituent fibers, but also the yarn structure characterized by geometrical arrangement of fibers in the yarn body. Although there are many studies related to analyzing the migratory properties of spun yarns, there are no studies available about predicting yarn migration parameters. Therefore, the main aim of this research is to introduce a new approach to predict migratory properties of different kinds of spun yarns, namely siro, solo, compact and conventional ring-spun yarns. To achieve the objectives of the research, general physical and mechanical properties of spun yarns together with existing standards were thoroughly studied. Spun yarn migratory properties were predicted using intelligent technique of artificial neural network (ANN). Results signified that the ANN models can predict precisely the yarn migratory properties on the basis of a series of yarn physical and mechanical properties.     

基于多元回归与人工神经网络的橡胶园树龄遥感反演研究

树龄是生产管理、森林生态系统研究如叶面积指数（LAI）反演,净初级生产量（NPP）估算的重要参数之一。以国营新盈农场为例,研究橡胶园树龄与美国陆地卫星TM影像之间的关系,分别应用多元回归和人工神经网络的方法建立了橡胶园树龄遥感反演模型。研究结果显示：1）TM波段和橡胶园树龄呈显著负相关,其中近红外（B4）、红外波段（B5）与树龄的相关系数最高,分别达到-0.70和-0.69;2）人工神经网络模型能克服建模数据非正态、非线性及共线的影响,能够明显的提高模型的预测精度,绝对预测误差（ε≤6年）的百分比为81.2％,远高于回归模型的69.2％; 3）由于橡胶树生长特性、自然灾害等因素的影响,多元回归模型和人工神经网络模型都存在估计偏差,即高估小于25龄的橡胶园树龄和低估对应老龄胶园的树龄。研究结果表明,利用人工神经网络的方法进行橡胶园生物物理参数遥感反演具有良好的应用前景。     

联合主成分分析与最小二乘支持向量机估测冬小麦叶面积指数

利用单一植被指数估测叶面积指数存在高光谱遥感丰富的波段信息易丢失和外界因素干扰大的缺点,但若将波段信息全部引入模型又会增加建模难度。为解决利用多波段信息估测叶面积指数的问题,利用主成分分析法(PCA)对光谱数据进行降维,之后将提取的主成分与最小二乘支持向量机(LS-SVM)模型相结合,构建冬小麦叶面积指数的高光谱估测模型,并与以4类植被指数作为LS-SVM输入参数建立的模型进行比较。结果表明,以主成分作为LS-SVM模型的输入参数建立的模型精度最高,模型检验集R2为0.71,检验集RMSE为0.56,估测结果较使用植被指数作为输入参数建立的模型精度高,稳定性好。该方法可为利用多波段信息进行大范围冬小麦叶面积指数的无损测定提供参考。     

An artificial neural network approach to prediction of the colorimetric values of the stripped cotton fabrics

This paper presents an artificial neural network (ANN) modeling by Levenberg-Marquardt (LM) algorithm for predicting the colorimetric values of the stripped cotton fabrics dyed using commercial reactive dyes. Achieving the expected efficiency in the application of stripping process is a very important aspect for the success of the reproduction. In the study, the predictions of L* and ΔE colorimetric values of stripped cotton samples for different stripping applications by artificial neural network are reported. We set up different network structures with different number of nodes in the hidden layer, the number of inputs and MSE of results as stopping criteria in order to get the best fitting model. According to the result of the best neural network models predicting L* and ΔE, we achieved 97 % of R for both of them. We are able to predict the L* value of the stripped samples using some working parameters as inputs with only 1.2 % error. We think that our results are very promising and the predictions of L* and ΔE values of stripped samples before applying any process are possible using the ANN model set up in the study, especially for L*.     

基于无人机多时相遥感影像的冬小麦产量估算

为高效准确地预测小麦产量,以浙江省冬小麦为研究对象,利用四旋翼无人机精灵4多光谱相机获取冬小麦5个关键生育时期（拔节期、孕穗期、抽穗期、灌浆期、成熟期）的冠层多光谱数据,选取多光谱相机的五个特征波段计算各生育时期的72个植被指数,分别通过逐步多元线性回归（SMLR）、偏最小二乘回归（PLSR）、BP神经网络（BPNN）、支持向量机（SVM）、随机森林（RF）构建不同生育时期的产量估算模型,最后采用决定系数（R）、均方根误差（RMSE）和相对误差（RE）对估算模型进行评价,筛选出最优估算模型。结果表明,基于随机森林建立的模型估算效果最优,SMLR、PLSR和SVM三种方法建立的模型估算效果接近。利用随机森林算法所建拔节期、孕穗期、抽穗期、灌浆期、成熟期模型的R、RMSE和RE分别为0.92、0.35、11%;0.93、0.33、10%;0.94、0.32、9%;0.92、0.36、9%;0.77、0.67、33%。模型验证时,抽穗期估算效果最好（R、RMSE和RE分别为0.91、0.35和15%）,拔节期、孕穗期、灌浆期估算效果接近且有很好的估算能力,成熟期估算精度最差（R、RMSE和RE分别为0.71、0.47和13%）。由此说明,结合机器学习算法和无人机多光谱提取的植被指数可以提高小麦产量估算效果。