Pet project for recognizing edible mushrooms

The meaning of the project: To learn how to work with the TensorFlow framework and Keras, this project was invented.

Project idea: Write a telegram bot that knows how to classify mushrooms. The convolutional neural network DenseNet201 was used for classification. The accuracy of 71.08% was obtained on the test sample

At the moment, the classifier recognizes 4 types of mushrooms:

1. Edible mushrooms (can be eaten without health risk)
2. Inedible (for one reason or another are not suitable for human consumption)
3. Poisonous (when consumed in normal doses, a person gets severe poisoning)
4. Hallucinogenic (contain hallucinogenic substances)

The classifier can also determine the absence of mushrooms in the photo.

To use it, you need to send a photo (not a file, but a photo) and get a predictable response.

Ways to improve the quality of the classifier:

1. Selection of better photos of mushrooms (incorrect photos are present in the dataset)
2. Magnification of photos for each type of mushroom (at the moment, there are 16 photos for one type of mushroom in the test sample)

Screenshots of the bot's work: