Pengaruh ekstraksi fitur warna pada akurasi klasifikasi sayuran menggunakan naive bayes

The manual process of identifying vegetable types by humans is often influenced by subjectivity, especially when distinguishing colors that have a high degree of similarity. This limitation of visual perception opens the possibility of misidentification. In the context of informatics engineering, the use of digital image processing technology is a relevant alternative to provide a more consistent and efficient identification process. Therefore, a system is needed that is able to recognize objects automatically by relying on the unique numerical characteristics of each image. This study aims to design and initiate the performance of a system in identifying vegetable types based on digital visual data. The main focus of the study is directed at analyzing the influence of chromaticity parameters on the level of object identification accuracy in five different vegetable categories. The research stages began with data collection in the form of 50 vegetable images consisting of cabbage, chayote, mustard greens, purple eggplant, and carrots. The system implementation was carried out using MATLAB R2017b software with the support of a Graphical User Interface (GUI). The image processing process includes several stages, namely preprocessing, segmentation to separate objects from the red background, and feature extraction in the form of average color values in RGB and HSV spaces. Furthermore, morphological parameters such as eccentricity and metrics were used to enhance the representation of object shape. The classification process was conducted using a probabilistic approach using the Naive Bayes algorithm. During the evaluation phase, the data was divided into 45 training images and 5 test images. Test results showed that the system achieved an average accuracy of 80.0%. Specifically, objects with contrasting visual characteristics, such as carrots and purple eggplants, were well recognized due to their distinct color dominance and relatively distinctive shapes. However, misclassification errors were still observed, particularly in the case of chayote, which was identified as mustard greens. This resulted in a precision value of 70.0% and a recall of 80.0%. Based on these results, it can be concluded that the chromaticity parameter plays a significant role in the identification process. The system performed optimally on objects with striking color differences, but still encountered challenges when handling objects with similar hue values. Overall, the developed system has demonstrated quite good performance and is suitable for use as an initial solution for automatic vegetable classification based on digital image processing.