MODEL DETEKSI JAMUR BERACUN MENGGUNAKAN ALGORITMA EFFICIENTDET
Muhammad Faqih H, Suprapto, Drs., M.Kom., Dr.
2023 | Skripsi | ILMU KOMPUTER
Poisonous mushrooms contain toxins that can harm the body when consumed. Based on data from several sources, poisoning cases of poisonous mushrooms occur every year. Most cases of poisoning occur due to errors when identifying the type of mushroom. The development of computer technology today allows the development of classification/detection methods for poisonous mushrooms using machine learning. The use of machine learning has the advantage of increasing the efficiency and accuracy of poisonous mushroom identification.
The purpose of the research conducted is to develop a model that can be used to detect poisonous mushrooms by applying the efficientdet machine learning algorithm. In this study, 5 efficientdet architectures were used, consisting of efficientdet D0 - efficientdet D4. Each architecture is divided into three main parts: backbone, BIFPN, and prediction box/class. The backbone section will perform feature extraction from input images at five levels of resolution. Features at five levels of resolution are then combined using BIFPN. The merged result is used to predict the bounding box and class.
The efficientdet model was trained using a 7494 data augmentation dataset consisting of 8 different mushroom classes. After testing using two ensemble boxes methods, the efficientdet D1 model with WBF produced the best performance with mAP values of 0,911, precision 0,898, and recall 0,90. These results are better when compared to the YoloV5 model on the same dataset which produces mAP 0.877, precision 861, and recall 0.802.
Kata Kunci : Jamur beracun, Deep learning, Deteksi objek, Efficientdet