MEMBANGUN MODEL MESIN LEARNING DENGAN CONVOLUTIONAL NEURAL NETWORK UNTUK DETEKSI PENYAKIT DAUN SELADA

Modern farming techniques such as hydroponics use nutrient mineral solutions as a growing medium, without the need for soil. Hydroponics allows plants such as lettuce to grow well. Lettuce, one of the vegetables that is easy to grow hydroponically, has good nutritional content for the body. According to the USDA National Nutrient Database, every 100 grams of lettuce contains 18 calories, 0.18 grams of fat, 1.38 grams of protein, and 2.89 grams of carbohydrates. Although the hydroponic system for growing lettuce offers several advantages, there are several obstacles that can affect the quality of the plant. Suboptimal crop yields can be influenced by many factors, such as less than optimal monitoring, high air temperatures, and nutrient imbalances in plants. In addition, lack of knowledge about plant conditions can also increase the likelihood of disease attacks. Therefore, other technologies are needed to assess the growth and development of lettuce based on leaf image data, and this can be processed through an artificial neural network or deep learning approach. The methodology of the convolutional neural network research process stages, begins with inputting the initial image data, then the image size resize process is compressed by reducing the basis matrix of each column and row of the image, resulting in a pattern of image data division, which will then be processed to produce an optimum classification process model. Model accuracy shows a significant increase during the first six epochs, indicating the model's ability to accurately classify training data. In epochs 7-10, the accuracy remains high, indicating the consistency of model performance. In the final stage (11-25), the accuracy approaches 100%, indicating the efficiency of the model in classifying training data. However, the validation results show variations in val loss and val accuracy in the early epochs (1-6), which are likely due to the lack of model generality and possible overfitting. However, in epochs 7-10, val loss and val accuracy are more stable, indicating better generalization on validation data.