Pertanian adalah sektor ekonomi utama di Asia Tenggara yang penting bagi pertumbuhan ekonomi, namun penyakit tanaman mengancam kualitas dan kuantitas produksi, serta keamanan pangan. Tanaman mangga populer di daerah tropis dan kaya akan nutrisi. Penyakit mangga biasanya dimulai dengan perubahan morfologi daun, dan identifikasi yang akurat sangat penting untuk pencegahan. Penyebab penyakit meliputi patogen, hama, gulma, serta faktor non-parasit seperti air, suhu, cahaya, dan nutrisi. Metode yang dipilih sebagai identidikasi penyakit daun mangga yaitu Convolutional Neural Network. Dalam penelitian ini dataset yang digunakan berasal dari platform Kaggle yang berjumlah 4000. Citra daun mangga dibagi menjadi 8 kelas Anthracnose, Bacterial Canker, Cutting Weevil, Die Back, Gall Midge, Healthy Leaf, Powdery Mildew, dan Sooty Mould. Arsitektur model yang digunakan dalam penelitian ini ada 5 yaitu VGG19, DenseNet201, InceptionV3, NASNetMobile, dan Xception dengan menggunakan Jupyter Anaconda sebagai alatnya, Uji coba dilakukan dengan memasaukkan 10, 15, dan 20 epoch dan batch size sebesar 32 dan 128. Hasil akurasi keseluruhan terbaik diperoleh dari model DenseNet201 dengan pixel 160x160 sebesar 99,06% dengan masukan batch size sebesar 32 dan epoch 10.

Agriculture is a major economic sector in Southeast Asia that is important for economic growth, but plant diseases threaten the quality and quantity of production, as well as food safety. Mango plants are popular in the tropics and are rich in nutrients. Mango diseases usually begin with changes in leaf morphology, and accurate identification is essential for prevention. Disease causes include pathogens, pests, weeds, as well as non-parasitic factors such as water, temperature, light, and nutrients. The method chosen for mango leaf disease identification is Convolutional Neural Network. In this study, the dataset used came from the Kaggle platform which amounted to 4000. Mango leaf images are divided into 8 classes Anthracnose, Bacterial Canker, Cutting Weevil, Die Back, Gall Midge, Healthy Leaf, Powdery Mildew, and Sooty Mold. There are 5 model architectures used in this study, namely VGG19, DenseNet201, InceptionV3, NASNetMobile, and Xception using Jupyter Anaconda as a tool, The trial was conducted by loading 10, 15, and 20 epochs and batch sizes of 32 and 128. The best overall accuracy results were obtained from the DenseNet201 model with 160x160 pixels of 99.06% with an input batch size of 32 and epoch 10.