This research aims to detect the physical characteristics of corn kernels and analyze images using a deep learning model. The data analysis based on the CRISP-DM framework which consists of six steps, business understanding, data understanding, data preparation, modelling, evaluation, and deployment. The business goal reduces the cost of the separation of abnormal corn kernels. The dataset comprises 1,800 images of corn kernels and divided equally between normal and abnormal corn kernels. The dataset was divided into three subsets: 1,000 images for training the deep learning model, 600 images for validation and 200 images for testing. The tools for analysis in this research are Jupyter Lab, Python, TensorFlow Keras, and Convolutional Neural Networks. The results revealed that the deep learning model achieved the accuracy rate of 99% in differentiating between normal and abnormal corn kernel images that is a highly effective model in this context.

翻译：本研究旨在通过深度学习模型检测玉米籽粒的物理特性并分析图像。数据分析基于CRISP-DM框架，该框架包含六个步骤：业务理解、数据理解、数据准备、建模、评估与部署。业务目标是降低异常玉米籽粒分选成本。数据集包含1,800张玉米籽粒图像，正常与异常籽粒图像各占半数。数据集被划分为三个子集：1,000张图像用于训练深度学习模型，600张用于验证，200张用于测试。本研究所用的分析工具包括Jupyter Lab、Python、TensorFlow Keras和卷积神经网络。结果表明，该深度学习模型在区分正常与异常玉米籽粒图像时达到了99%的准确率，在此应用场景中展现出极高的有效性。