SAPUTRA, MUHAMMAD DONI (2026) KLASIFIKASI MULTI KELAS PENYAKIT DAUN STROBERI MENGGUNAKAN CONVOLUTIONAL NEURAL NETWORK DENGAN ARSITEKTUR MOBILENETV2. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (HAL COVER)
COVER.pdf Download (559kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
BAB I.pdf Restricted to Registered users only Download (245kB) |
|
|
Text (BAB II)
BAB II.pdf Restricted to Registered users only Download (468kB) |
|
|
Text (BAB III)
BAB III.pdf Restricted to Registered users only Download (398kB) |
|
|
Text (BAB IV)
BAB IV.pdf Restricted to Registered users only Download (656kB) |
|
|
Text (BAB V)
BAB V.pdf Restricted to Registered users only Download (202kB) |
|
|
Text (DAFTAR PUSTAKA)
DAFTAR PUSTAKA.pdf Restricted to Registered users only Download (236kB) |
|
|
Text (LAMPIRAN)
LAMPIRAN.pdf Restricted to Registered users only Download (436kB) |
Abstract
Strawberry leaf diseases, such as Angular Leaf Spot and Leaf Scorch, often become obstacles that impact quality and production yields if not detected early. Generally, farmers identify plant diseases through manual visual observation. However, this method has several weaknesses that can affect the identification results, such as the process is relatively time-consuming, and is susceptible to subjectivity and limited knowledge in recognizing disease symptoms accurately. This study aims to build an automatic classification system to identify strawberry leaf disease conditions segmented into three domains: Angular Leaf Spot, Healthy, and Leaf Scorch, using deep learning techniques. The MobileNetV2 architecture, which uses pre-trained weights from the ImageNet dataset as feature extractors and applies a transfer learning approach, is used to build the model. Meanwhile, additional classification layers such as Global Average Pooling, Dense, and Dropout are retrained according to the research needs. The model learning process is implemented by utilizing 2,092 training data images and 261 validation data images for 10 epochs, with the Adam optimizer and ReduceLROnPlateau callback to maintain the stability of the optimization process. With consistently high precision, recall, and f1-score values across all classes, ranging from 0.98 to 1.00, the model achieved a 99% accuracy rate based on testing on 262 test data images. The model presented a stable confidence level across varying image conditions. These results indicate that the MobileNetV2-based transfer learning technique is useful for classifying strawberry leaf diseases with a high level of accuracy and relatively low computational requirements; therefore, this technique can be further developed as an early detection tool in the field. Keywords: MobileNetV2, transfer learning, image classification, strawberry leaf disease, deep learning Penyakit daun tanaman stroberi, seperti Angular Leaf Spot dan Leaf Scorch, kerap menjadi kendala yang berdampak pada penurunan kualitas maupun hasil produksi apabila tidak terdeteksi secara dini. Pada umumnya, petani mengidentifikasi penyakit tanaman melalui pengamatan visual secara manual. Namun, metode tersebut memiliki berbagai kelemahan yang dapat mempengaruhi hasil identifikasi, seperti proses yang memakan waktu relatif lama, serta rentan terhadap unsur subjektivitas dan keterbatasan pengetahuan dalam mengenali gejala penyakit secara tepat. Penelitian ini bertujuan membangun sistem klasifikasi otomatis untuk mengidentifikasi kondisi penyakit daun stroberi tersegmentasi ke dalam tiga domain, yaitu Angular Leaf Spot, Healthy, dan Leaf Scorch, dengan memanfaatkan teknik deep learning. Arsitektur MobileNetV2, yang menggunakan bobot pra�terlatih dari dataset ImageNet sebagai ekstraktor fitur dan menerapkan pendekatan pembelajaran transfer, digunakan untuk membangun model tersebut. Sedangkan lapisan klasifikasi tambahan berupa Global Average Pooling, Dense, dan Dropout dilatih kembali sesuai kebutuhan penelitian. Proses pembelajaran model diterapkan melalui pemanfaatan 2.092 citra data latih dan 261 citra data validasi selama 10 epoch, dengan optimizer Adam serta callback ReduceLROnPlateau untuk menjaga kestabilan proses optimasi. Dengan nilai presisi, recall, dan f1-score yang konsisten tinggi di semua kelas, berkisar antara 0,98 hingga 1,00, model ini berhasil mencapai tingkat akurasi 99% berdasarkan pengujian pada 262 gambar data uji. Model menyajikan Tingkat keyakinan yang stabil pada berbagai kondisi citra yang bervariasi. Hasil ini menunjukkan bahwa teknik transfer learning berbasis MobileNetV2 bermanfaat untuk mengklasifikasikan penyakit daun stroberi dengan tingkat akurasi tinggi dan kebutuhan komputasi yang relatif rendah; oleh karena itu, teknik ini dapat dikembangkan lebih lanjut sebagai alat deteksi dini di lapangan. Kata kunci: MobileNetV2, transfer learning, klasifikasi citra, penyakit daun stroberi, deep learning
Actions (login required)
 |
View Item |
