Modern technologies have improved their application in field of agriculture in order to improve production. Plant diseases are harmful to plant growth, which leads to reduced quality and quantity of crop. Early identification of plant disease will reduce the loss of the crop productivity. So, it is necessary to identify and diagnose the disease at an early stage before it spreads to the entire field. In this chapter, the proposed model uses VGG16 with attention mechanism for leaf disease classification. This model makes use of convolution neural network which consist of convolution block, max pool layer, and fully connected layer with softmax as an activation function. The proposed approach integrates CNN with attention mechanism to focus more on the diseased part of leaf and increase the classification accuracy. The proposed model design is a novel deep learning model to perform the fine tuning in the classification of nine different type of tomato plant disease.
TopIntroduction
Unpredictable climate change, without any proper irrigation technique, lack of knowledge in using modern technologies and use of pesticides have caused an imbalance in farming which has caused improper cultivation and unhealthy crops thereby threatening food security. Plant pathogens will lead to crop loss. Due to animals, weeds, pathogens it is estimated to reduce agricultural production between 20% to 40% globally. Agriculture is one the main sources of income in India it struggles to support rapidly growing population. It is estimated that in India smallholder farmers will generate around 80 percent of agricultural production and due to pests and plant disease there is more than 50 percent of yield loss. Plant diseases reduce the quality and production of food crops. To overcome this we need large verified dataset of images which includes healthy and diseased leaf images of all type of crop plants. Using this dataset develop an accurate image classifier which classifies the type of disease at initial stage. Such dataset was not available until recent and smaller dataset will not give accurate classification. So, PlantVillage project was initiated to benefit farmers with early disease detection and started collecting thousands of images of all types of crops. This chapter focuses on automatic plant disease detection as a topic of discussion. The study demonstrates the technical feasibility of deep learning to enable automatic tomato plant disease detection through tomato leaf images. The dataset consist of both diseased and healthy leaf images. In this chapter, deep convolution neural network is used to categorize the plant leaf as healthy or diseased. Various diseases of tomato leaf is identified like tomato_mosaic, lateblight, yellow curved, septoria Leaf Spot, healthy, bacterial spot with total 21071 images . These diseases can be identified at initial stage using CNN and farmers can use any infection control tools to stop spreading disease to other plants and also solve pest problems while minimizing the risks of human and environment. Some of the challenges identified are as follows.
- •
Quality of the leaf image
- •
Larger dataset
- •
Image denoising
- •
Segmenting exact spot of disease in leaf
- •
Splitting the training and testing samples from original images
- •
Feature extraction like color, size, texture and shape from image
- •
Recognizing different type of disease from plant leaves
Leaf diseases are threat to the crop production and farmer’s economy reduces immensely with increase in spread of crop disease. Conventional methods of detecting crop diseases require a great deal of knowledge and expertise. Further, these methods can be expensive, time consuming, and even ineffective in some cases. These challenges can be solved using deep learning techniques which are having more potential to identify various diseases. This motivates the work to focus mainly on leaf disease detection and classify the disease at early stage with great accuracy and prevent farmers from huge loss. The study uses tomato leaf for analysis as tomato is considered most profitable crop because it is grown four times in a year. Plant disease is quite natural because they cannot withstand light intensity, high humidity, temperature etc., it affects plant pigmentation, fruit color and fruit quality and temperature below 10 degree Celsius and above 38 degree Celsius affects plant tissue and plant. Due to all these environmental conditions plants are susceptible to diseases like fungus, bacteria, viruses etc., Early detection of plant disease can help in increasing yield by preventing from disease. CNN improves the classification accuracy in many fields, including agriculture. So, deep convolution neural network is used for demonstration. The model proposed hybrid deep learning model, which consists of transfer learning, attention model, and dropout operation, is proposed along with a CNN-based VGG16 classifier to detect and classify diseases associated with plants. Pretrained imageNet weights were used for image preprocessing. Pre-trained images are trained on millions of images with 1000 different categories, which reduce the time for training new images. Further, all images were 3-dimensional, i.e., they were measured in terms of height, width, and channels (RGB). Objectives of the proposed system are: