A MULTI-DATASET DEEP LEARNING APPROACH TO BRINGEL LEAF CERCOSPORA DISEASE DETECTION

Abstract

Deep learning became a platform for the most accurate and precise detection of plant diseases. The proposed work is to find a grow crops in the local area (Lower dir.)Khyber Pakhtoon khwa, and lists the effects of these crops diseases. The next step is to crop diseases and the re-training model. After the new recognition, accuracy is obtained again crop diseases. Local crop data set is not open, we need to add them, so we list the vegetables, and fruit crops crop growth in the local field.   In the existing system it is mainly, a village began using plant data set. They analyzed 54,306 images of plant leaves, with their assigned class label 38 to spread. Each class labels are crop diseases, plant leaf crop disease only image they do try to predict given right. They adjust the image to 256 × 256 pixels, we perform optimization of these two models and forecasts for these reduced image. At present, the factory village set uses which are public; we added a local data set. It is a bit like classification label for Bringel_leaf_cercospora. This network has 62.3 million parameters in Crop Disease Net, RELU forward pass and one billion computing needs we can see convolution layer, 6% of all parameters, calculation of the consumption of 95%.  We use Alex net by a convolution layer 5, Followed via 3 fully related Layers (FC), and subsequently finishing with a soft Max layer. Google Net architecture is very deep and the wider framework, and then AlexNet. It has 22 floors. AlexNet while still having a relatively low ratio of number of network parameters (5 Parametric) (60000000 parameters). The total number of modules in use since version Google Net architecture of nine, we used in our experiments.They have a total of 60 forms of the experimental setup, which range the following parameters: Tensor Flow is an open source software library of excessive-performance numerical computing. Keras is a high-level neural network API, which is written in Python language and capable of running on top of Tensor Flow, Theano or CNTK.Anaconda is Python and scientific computing .We had also implemented a validation set. The only test has been made to improve the precision to find the best model, and after that, we deploy lives. The extracted features are blended with 100 times the neural network we get about. A comparison of the Alex Net and Crop Disease Net architectures. From the comparison, we get 98% accuracy for the Crop Disease Net architecture and we get the accuracy of 97 % for Alex net. Therefore, the Crop Disease Net architecture performs better than Alex net to identify disease sills from leaf images. After that, accuracy measures in test model and final stage will be model deployment stage.