KLASIFIKASI KEMATANGAN CABAI BERDASARKAN CITRA DENGAN PENGHAPUSAN LATAR MENGGUNAKAN GRABCUT DAN U2-NET BERBASIS CNN

Chili peppers (Capsicum spp.) are one of Indonesia’s most important horticultural commodities with significant economic value. The ripeness level of chili peppers greatly affects their quality, price, and shelf life, making accurate classification essential. Manual assessment by humans is often subjective and inconsistent; therefore, this study proposes an automated approach using a Convolutional Neural Network (CNN) to classify chili ripeness levels. The dataset consists of 1,964 chili images categorized into five classes: unripe, half-ripe, ripe, half-rotten ripe, and rotten ripe. The study also analyzes the effect of background removal using two segmentation methods GrabCut and U²-Net on CNN performance. Results show that both methods successfully removed the background, with U²-Net producing cleaner and more consistent segmentation. The CNN model achieved the highest accuracy of 96% on the original dataset (without background removal), followed by U²-Net at 92% and GrabCut at 89%. These findings indicate that background removal does not always improve model accuracy, as background information can sometimes aid the model in recognizing visual features of the chili. Therefore, the original dataset is recommended for chili ripeness classification, while U²-Net serves as a reliable alternative for complex background conditions.