ABSTRACT Considering that damages and forces to the fruit cause quantitative and qualitative changes in the fruit, in this study, the effects of three levels of loading force (wide and thin edges) (15, 30, and 45 N), 2 fixed positions on the Instron fixed jaw (vertical and horizontal), and 3 storage periods on Hayward kiwi were investigated. Experiments were analyzed as a completely randomized factorial design using SAS statistical software and data were analyzed for prediction using a multilayer perceptron artificial neural network. Statistical results showed that weight, volume, and density of kiwi fruit were decreased for loading of wide and thin edges, and according to the results, it can be concluded that weight loss in wide edge loading was more than loading of thin edges. Also, the weight, volume, and density of the fruit decreased significantly when the fruit was extensively loaded. For neural networks the best R value for weight, volume, and density were 0.9992, 0.99840, and 0.997, respectively, and for RMSE which should be the lowest among the networks, 0.22584, 3091.13 and 0.0049, respectively. Overall, it can be stated that the neural network was capable of predicting weight, volume, and density for both types of loading. But for the wide edge, equivalent, geometric, and arithmetic diameters, and for the thin edge of the aspect ratio and rationality coefficient have had a far greater impact on artificial neural network improvement and data prediction. In brief, for loading the thin edge of the network with loading force input, storage period, loading direction, spherical coefficient, spherical coefficient, aspect ratio coefficient, length, width, and thickness (network 2) and for loading wide edge, loading force, storage period, loading direction, equivalent diameter, geometric diameter, arithmetic diameter, length, width, and thickness were the best in terms of accuracy and error.

中文翻译：

---

猕猴桃不同装载、贮藏过程中物理性质变化的研究及人工神经网络建模

摘要 考虑到果实的损伤和受力会引起果实的量变和质变，本研究采用三个水平的加载力（宽边和细边）（15、30 和 45 N）、2 个固定位置对水果的影响。对 Instron 固定钳口（垂直和水平）以及 Hayward 猕猴桃的 3 个储存期进行了研究。使用 SAS 统计软件将实验分析为完全随机的因子设计，并使用多层感知器人工神经网络分析数据以进行预测。统计结果表明，猕猴桃的重量、体积和密度随着宽边和薄边的装载而降低，根据结果可以得出结论，宽边装载的重量损失大于薄边装载。此外，重量、体积、当果实大量装载时，果实的密度显着降低。对于神经网络，重量、体积和密度的最佳 R 值分别为 0.9992、0.99840 和 0.997，而对于网络中最低的 RMSE，分别为 0.22584、3091.13 和 0.0049。总的来说，可以说神经网络能够预测两种负载类型的重量、体积和密度。但对于宽边、等效、几何和算术直径，对于细边的纵横比和合理系数对人工神经网络改进和数据预测的影响要大得多。简而言之，对于加载力输入、存储周期、加载方向、球面系数、球面系数、纵横比系数的网络细边加载，