Refining Mushroom Image Recognition Using Deep Learning and Image Similarity Distance

This study investigates the characteristics of images color, texture, and shape to identify specific types of fungi. The various mushroom species in our environment, their suitability for human consumption requires thorough research. The primary goal of this research was to classify mushrooms that are safe to eat and may contain potent anti-cancer agents. We utilized several Convolutional Neural Network (CNN) models to extract the characteristics of mushroom images. These models include DenseNet169, InceptionResnetV2, Xception, MobileNetV2, NasNetmobile, ResNet50V2, and InceptionV3. The retrieval process involved comparing the similarity between the feature extraction results of the query image and the dataset image. To measure the similarity of the distance between the query mushroom image and the dataset, we used several methods such as Euclidean, Cosine, Makowski, and Manhattan. After conducting numerous experiments to identify optimal mushroom images, the DenseNet169 and MobileNetV2 models successfully identified 15 classes of mushrooms with an accuracy level of 100%, using similarity distance methods, namely Cosine and Euclidean. However, it's worth noting that the DenseNet169 model exhibited a lower accuracy level in recognizing Straw Mushrooms due to their unique shape. The use of CNN models in extracting mushroom image characteristics has proven effective, paving the way for more advanced and accurate identification methods in the future. This study emphasizes the importance of image characteristics in identifying specific types of fungi, particularly those safe for human consumption.