@article { author = {Zandi, M. and Ganjloo, A. and Bimakr, M.}, title = {Applying Adaptive Neuro-Fuzzy Inference System and Artificial Neural Network to the Prediction of Quality changes of Hawthorn Fruit (Crataegus pinnatifida) during Various Storage Conditions}, journal = {Journal of Agricultural Machinery}, volume = {11}, number = {2}, pages = {343-357}, year = {2021}, publisher = {Ferdowsi University of Mashhad}, issn = {2228-6829}, eissn = {2423-3943}, doi = {10.22067/jam.v11i2.86654}, abstract = {IntroductionIn recent decades, artificial intelligence systems were employed for developing predictive models to estimate and predict many agriculture processes. Neural networks have the capability of identifying complex nonlinear systems with their own high learning ability. Artificial Neural Networks as a modern approach has successfully been used to solve an extensive variety of problems in the science and engineering, exclusively for some space where the conventional modeling procedure fail. A well-trained Artificial Neural Networks can be used as a predictive model for a special use, which is a data processing system inspired by biological neural system. The short storage life of hawthorn fruit and its high susceptibility to water loss and browning are the main factors limiting its marketability. So, it is important to evaluate parameters that affected the hawthorn quality. An adaptive neuro-fuzzy inference system or adaptive network-based fuzzy inference system (ANFIS) is a kind of artificial neural network that is based on Takagi-Sugeno fuzzy inference system. To estimate changes in fruit quality as a function of storage conditions, the evolution of certain quality-indicative properties such as color, firmness or weight can be used to provide related information on the quality grade of the product stored. Measurement of these parameters is an expensive and time-consuming process. Therefore, parameter prediction due to affecting factors will be more useful. In this study, the physicochemical properties of hawthorn fruit during various storage was predicted using artificial neural networks method. Hawthorn (Crataegus pinnatifida), belonging to the Rosaceae family, consists of small trees and shrubs. The color of the ripe fruit ranges from yellow, through green to red, and on to dark purple. Hawthorn is one of the most widely consumed horticultural products, either in fresh or processed form. It is also an important component of many processed food products because of its excellent flavor, attractive color and high content of many macro- and micro-nutrients.Materials and MethodsThe purpose of this study was a prediction of color, physical and mechanical properties of hawthorn fruit (Crataegus pinnatifida) during storage condition using artificial neural networks (ANNs) and adaptive network-based fuzzy inference system (ANFIS). Experimental data obtained from fruit storage, were used for training and testing the network. In the present research, artificial neural networks were used for modeling the relationship between physicochemical properties and color attributes with different storage time. Several criteria such as training algorithm, learning function, number of hidden layers, number of neurons in each hidden layer and activation function were given to improve the performance of the artificial neural networks. The total number of hidden layers and the number of neurons in each hidden layer were chosen by trial and error. The network’s inputs include storage time, hawthorn moisture content and storage temperature and the network’s output were the values of the physicochemical and color properties. The training rules were Momentum and Levenberg-Marquardt. The transfer functions were TanhAxon and SigmoidAxon.Results and DiscussionTo predict the weight loss and firmness multilayer perceptron network with the momentum learning algorithm, topologies of 3-15-5-1 and 3-8-5-1 with R2=0.9938 and 0.9953 were optimal arrangement, respectively. The optimal topologies for color change, hue, Chroma were 3-9-7-1 (R2=0.9421), 3-9-3-1 (R2=0.9947) and 3-7-1 (R2=0.9535) respectively, with momentum learning algorithm and TanhAxon activation function. The best network for ripening index prediction was Multilayer perceptron network with the TanhAxon activation function, Levenberg-Marquardt Levenberg-Marquardt learning algorithm, topology of 3-5-1-1 and R2=0.9956.Conclusion Three factors including firmness, total soluble solids and titratable acidity were considered for ripening index calculation during fruits storage condition. Momentum and Levenberg-Marquardt learning algorithms with SigmoidAxon and TanhAxon activation functions were used for training the patterns. Results indicated artificial neural networks to be accurate and versatile and they predicted the quality changes in hawthorn fruits. The outcomes of this study provide additional and useful information for hawthorn fruits storage conditions.}, keywords = {Adaptive neuro-fuzzy inference system,Artificial neural network,Hawthorn,Multilayer perceptron}, title_fa = {به‌کارگیری سیستم استنتاج عصبی-فازی سازگار و شبکه‌های عصبی مصنوعی در پیش‌بینی و مدل‌سازی تغییرات کیفی زالزالک (Crataegus pinnatifida) طی شرایط مختلف انبارمانی}, abstract_fa = {در دهه‌های اخیر، از سیستم‌های هوش مصنوعی برای ایجاد مدل‌های پیش‌بینی جهت تخمین و پیش‌بینی بسیاری از فرآیندهای کشاورزی استفاده شده است. در این مطالعه، خصوصیات فیزیکی و شیمیایی میوه زالزالک طی نگهداری در شرایط مختلف با استفاده از شبکه‌های عصبی مصنوعی و سیستم استنتاج عصبی-فازی سازگار پیش‌بینی گردید. از داده‌های تجربی حاصل از نگهداری میوه، برای آموزش و آزمایش این شبکه‌ها استفاده شد. تعداد کل لایه‌های پنهان و تعداد نورون در هر لایه پنهان به روش سعی و خطا انتخاب گردید. شبکه عصبی و سیستم استنتاج عصبی-فازی سازگار طراحی شده دارای ورودی شامل زمان نگهداری، رطوبت اولیه و دمای نگهداری و یک متغیر در لایه‌های خروجی (WL، F،c* ، *h و RPI) بود. مقادیر R2 بالا و RMSE کم گویای کارایی بالای مدل شبکه عصبی مصنوعی و سیستم استنتاج عصبی-فازی سازگار در پیش‌بینی خصوصیات کیفی زالزالک طی فرآیند نگهداری می‌باشد. نتایج نشان داد که شبکه عصبی پرسپترون چندلایه با الگوریتم یادگیری مومنتوم و تابع آستانه‌ای تان‌اکسون بهترین شبکه برای پیش‌بینی خصوصیات کیفی زالزالک در شرایط مختلف بود. نتایج مدل‌سازی با انفیس نشان داد که توابع عضویت ذوزنقه‌ای و گوسی بهترین عملکرد را به‌ترتیب در پیش‌بینی پارامترهای رنگی و فیزیکی داشت. با مقایسه نتایج حاصل از مدل‌سازی با شبکه عصبی مصنوعی و انفیس، تفاوت زیادی از نظر دقت و کارایی در پیش‌بینی مشاهده نشد، اگرچه شاخص RMSE در مدل‌سازی با کمک انفیس کمتر از شبکه عصبی مصنوعی بود که خود نمایان‌گر دقت بالاتر آن می‌باشد.}, keywords_fa = {زالزالک,سیستم استنتاج عصبی-فازی سازگار,شبکه پرسپترون چند لایه‌ای,شبکه عصبی مصنوعی}, url = {https://jame.um.ac.ir/article_34830.html}, eprint = {https://jame.um.ac.ir/article_34830_8f368336f7d0fa0a1e84354f53cf0f39.pdf} }