METODE CNN UNTUK KLASIFIKASI KUALITAS TOMAT DENGAN TEKNIK CONTRAST ENHANCEMENT PADA CITRA

Tomato (Solanum lycopersicum) is a high-value horticultural commodity whose quality is often assessed manually through visual observation, which is prone to inaccuracy, inconsistency, and inefficiency for industrial-scale operations. The main problem of this study is the need for an automatic, objective, and accurate classification system to determine tomato quality. This research aims to develop a tomato quality classification model using Convolutional Neural Network (CNN) with Contrast Limited Adaptive Histogram Equalization (CLAHE) as a preprocessing technique to enhance image quality. The methodology includes collecting a tomato image dataset (2,438 images, three classes: ripe, unripe, rotten), applying CLAHE, splitting data (70% training, 15% validation, 15% testing), training the CNN model, and evaluating performance using accuracy, precision, recall, and F1-score metrics. The results show that the CNN model with CLAHE preprocessing achieved the highest accuracy of 91%, higher than the model without CLAHE (90%). Significant improvements were observed in the recall of the rotten class (0.79→0.86) and the precision of the ripe class (0.84→0.88), however, there was a slight decrease in rotten class precision and mature class recall due to visual similarities between classes. In conclusion, the combination of CNN and CLAHE can improve accuracy and stability in tomato quality classification, although visual evaluation remains necessary to address misclassifications in visually similar classes.