Tanaman strawberry merupakan salah satu komoditas hortikultura yang rentan terhadap serangan penyakit yang dapat menurunkan kualitas serta kuantitas hasil panen. Proses identifikasi penyakit secara manual masih bergantung pada pengamatan visual sehingga menyebabkan kesalahan diagnosis, keterlambatan penanganan, serta meningkatkan risiko penyebaran penyakit. Penelitian ini bertujuan mengembangkan aplikasi deteksi penyakit tanaman strawberry berbasis Convolutional Neural Network (CNN) menggunakan arsitektur Xception untuk membantu proses identifikasi penyakit secara cepat dan akurat. Dataset yang digunakan terdiri dari 4.000 citra daun strawberry yang terbagi ke dalam empat kelas, yaitu healthy, leaf blight, leaf scorch, dan leaf spot. Dataset dibagi menjadi data training sebanyak 3.200 citra, validation 400 citra, dan testing 400 citra dengan rasio 80:10:10. Tahapan pre-processing meliputi resizing citra menjadi 224×224 piksel, normalisasi data, serta augmentasi citra berupa random flip horizontal, random rotation 0.1, random zoom 0.1, dan random contrast 0.1 untuk meningkatkan variasi data pelatihan. Model dibangun menggunakan transfer learning Xception dengan bobot ImageNet, GlobalAveragePooling2D, dense layer 256 neuron, dropout 0.3, dense layer 128 neuron, serta output deteksi dan bounding box. Proses pelatihan menggunakan optimizer Adam dengan learning rate 0.0001, batch size 16, dan epoch maksimum 50. Hasil penelitian menunjukkan model memiliki performa yang sangat baik pada seluruh kelas. Nilai precision yang diperoleh adalah 98% pada kelas healthy, 99% pada kelas leaf blight, 100% pada kelas leaf scorch, dan 100% pada kelas leaf spot. Nilai recall masing-masing kelas mencapai 99%, 99%, 100%, dan 98%, sedangkan nilai F1-score sebesar 99% pada kelas healthy, 99% pada leaf blight, 100% pada leaf scorch, dan 99% pada leaf spot. Secara keseluruhan, model mencapai akurasi (accuracy) sebesar 99%, yang menunjukkan kemampuan yang sangat baik dalam mendeteksi penyakit tanaman strawberry.

Strawberry plants are a horticultural crop that is susceptible to disease outbreaks, which can reduce both the quality and quantity of the harvest. Manual disease identification still relies on visual observation, leading to misdiagnosis, delayed treatment, and an increased risk of disease spread. This study aims to develop a Convolutional Neural Network (CNN)-based strawberry plant disease detection application using the Xception architecture to facilitate rapid and accurate disease identification. The dataset consists of 4,000 strawberry leaf images divided into four classes: healthy, leaf blight, leaf scorch, and leaf spot. The dataset is split into 3,200 training images, 400 validation images, and 400 testing images with a ratio of 80:10:10. The pre-processing steps include resizing the images to 224×224 pixels, data normalization, and image augmentation in the form of random horizontal flipping, random rotation of 0.1, random zoom of 0.1, and random contrast of 0.1 to increase the variation in the training data. The model was built using Xception transfer learning with ImageNet weights, GlobalAveragePooling2D, a 256-neuron dense layer, 0.3 dropout, a 128-neuron dense layer, and detection and bounding box outputs. The training process used the Adam optimizer with a learning rate of 0.0001, a batch size of 16, and a maximum of 50 epochs. The results of the study show that the model performs very well across all classes. The precision values obtained were 98% for the healthy class, 99% for the leaf blight class, 100% for the leaf scorch class, and 100% for the leaf spot class. The recall values for each class were 99%, 99%, 100%, and 98%, respectively, while the F1-scores were 99% for the healthy class, 99% for the leaf blight class, 100% for the leaf scorch class, and 99% for the leaf spot class. Overall, the model achieved an accuracy of 99%, demonstrating excellent performance in detecting strawberry plant diseases.