Agriculture is one of the major sectors in our society and from the medieval times, it is the sector that we have dwelled upon. About 60-70% of Indian population depends on the agricultural industry. Across the world, loss faced by the crop due to numerous factors like weeds, diseases and arthropods have increased to an alarming rate of about 34.9% in 1965 to about 42.1% in the late 1990s. Potato plants face various infections due to bacteria and fungi. Fungal diseases infecting plants are early blight and late blight. Our paper presents a CNN model and an algorithm to detect such diseases through the leaf of the crop. The model that we have created is trained to analyse and understand a diseased leaf and thereafter recognise the disease of the leaf. We are using InceptionV3 algorithm.

For any civilization, agriculture is one of the building block in the society. For many years India is one of the major countries involved in the agricultural sector . With a stellar history in agriculture, this industry forms an important part of the economy. Among the various crops grown in India, potato is the most in-demand crop. In India’s diverse range of subtropical climate, growing vegetables like potato yields fruitful results. 

Detecting diseases in plants in early stages is one of the major concerns in Agriculture field. So, there are many researchers who are already working on the issue of detecting plant diseases and diagnosing them some of the results of the research done in detecting plant diseases are as follows: 


We have used convolutional neural networks for the purpose of classification. After acquiring features from inception v3  deep neural networks are used to train the model to detect presence of early blight and late blight in the images. During classification Adam optimiser is used to reduce the training time and quickly converge the loss . Adam  is an extension of SGD that is now widely used for computer vision. For classification in different labels Softmax activation function is used. The softmax activation function is a function that converts a vector of n real values to a vector of n real values that totals up to 1. The input values can be any number but the softmax transforms them into values between 0 and 1, so that they can be interpreted as probabilities.

The CNN model contains an input layer where all images are fed in the start. After which the images are fed to the inception v3 architecture to extract features. These feature extracted images are sent to deep neural nets that classify our images on pre-trained knowledge. Finally, the images are sent to the output layer. The model has a couple of layers that are first created and after that they are compiled together using functions from the TensorFlow library.

CNN Layer

The model that we have proposed in this paper has used the plant village dataset [1] which has about 1000 leaf images of early blight and 152 images of healthy plants. For this model, the dataset has been divided into 2 parts that are the training set and the test set. Training set comprises of 80% of the dataset while the test set comprises of 20% of the dataset. The pre-trained model used on this dataset for feature extraction is Inception V3. Here for classification our CNN model provides an accuracy of about 90% on the basis of the training and testing done on it.

In this research, we have made a CNN model with the help of Inception V3 architecture and Adam Optimiser to diagnose and classify disease of potato plants such as early and late blight where we achieved an accuracy of 90% over the test dataset in classification. With the help of our model, a farmer can build a computer setup from where he can monitor the plants health issues efficiently, enhance his crop yield and detect and diagnose diseases in the early stages itself.