Sistem Klasifikasi Keanekaragaman Tanaman Pangan Menggunakan Transfer Learning Pendekatan CNN dan Model Arsitektur EfficientNetB7

Plant species identification is a crucial aspect in agriculture and forestry, significantly impacting food production, environmental conservation, and scientific research. The difficulty in identifying plant species can be caused by several factors, such as high morphological diversity, similarities between species, and changes in plant morphology due to different environmental conditions. This study uses a deep learning approach with the EfficientNetB7 architecture to solve the problem of plant identification. The dataset used consists of 30,000 images representing 30 plant species, each with 1,000 images. The model was trained using transfer learning techniques, tested on two scenarios classification with 4 plant classes and 30 plant classes. Results showed an accuracy of 97% with a loss of 0.24 for 4 classes, and an accuracy of 85% with a loss of 1.1 for 30 classes. The higher loss value in the scenario with 30 classes was due to the increased complexity and greater diversity of data. The evaluation results showed that the EfficientNetB7 was effective in classifying plant species with a high level of accuracy. It’s expected that model can be implemented to improve efficiency in plant maintenance and management. Convolutional Neural Network (CNN) architecture greatly influences the results of image classification. CNN is generally divided into two stages feature extraction using convolution layers and classification using artificial neural networks. The sixth CNN succeeded in achieving the highest accuracy in batik motifs, which was 87.83%. This model was good performance on precision and recall metrics.