Agriculture is one of the important sources of earning worldwide. With the rapid expansion of the human population and food security for all, the agriculture sector needs to be boosted to increase the yield. Agriculture is the prime source of livelihood in India for more than 50% of the total population. As per Indian agriculture and allied industries industry report, agriculture is one of the major contributors in gross value. Agricultural crops suffer heavy losses due to insect damage and plant diseases. Worldwide, out of the crop losses, major losses are caused by plant pests. In this chapter, various image pre-processing methods and the need of pre-preprocessing are discussed in detail. For image classification, TensorFlow deep neural network is presented. Deep learning model is used for automatic and early detection of paddy pests. Early detection of the pests will aid farmers in adopting necessary preventive measures. Multiple ways to reduce overfitting during model training are also suggested.
TopImage Variations And Pre-Processing Methods
The image classification system must take into consideration the following variations in the images:
Viewpoint Variation: A single instance of an object can be photographed or viewed in many ways with different camera positions (Chu et al., 2019). The image of an object captured in different angles must be labeled to the same class.
Scale Variation: Images can have different sizes. During pre-processing, the images are resized for uniformity.
Deformation: Many objects of interest are not rigid bodies and can be deformed. Deformed object images must be labeled correctly.
Occlusion: The objects of interest can be hidden behind another type of object in the image and only a small portion of an object is visible.
Illumination Conditions: The effects of illumination are drastic on the pixel level. Image classification algorithm must be able to handle changes in illumination.
Background Clutter: When the object of interest and background of the image is similar. The objects of interest may blend into the background. This makes the image classification task hard.
Intra-Class Variation: When objects belonging to one category or class have variations. For example plant leaf images belonging to one type of disease has variation in symptoms.
Images acquired from different devices are not directly provided as input to a machine learning algorithm, instead, images are pre-processed to enhance the image dataset. Pre-processing involves noise reduction, brightness, and contrast enhancements. Thus the aim of image pre-processing is to remove noise or unwanted data and enhance the image features, the consistent physical geometry of the image, image resizing, image augmentation, etc (Elgendi et al., 2021). A consistent and enhanced image dataset will improve the accuracy of the model. Different image pre-processing techniques to enhance the pictorial information for interpretation and analysis are