This dataset is about tomato leaf disease and I think it is a sample of some datasets, This dataset has 10 classes and each class contains 1000 images for training and 100 for validation, but at first, I separated 100 images from the train and created test folder and move images to it. The dataset is available in this link.
This model has a simple Convolutional Neural Network that we implement with TensorFlow and it gets images with the size of 1281283. We train data with just 100 epochs and have a 99% accuracy in training and 97% in testing and it just needs to be trained with more epochs. The model size is about 5 megabytes and if you convert it to tf-lite the size of the model will decrease you can use it in embedded systems such as Raspberry Pi and Jetson Nano for real-time and this is the novelty of this project than similar projects. According to my experience, the inference of the model in Raspberry Pi 4 is less than 200 milliseconds. In Colab I process 64 images in 120 milliseconds and with simple division you find out that Colab processes images for each image in 2milisecond, So according to this result you can use it in many embedded devices. The tflite model size is less than 500kb and you can use it in every Raspberry Pi that supports tensorflow or tflite_runtime and processes images in real-time.