Mobile Application Based Seed Counting Analysis Using Deep-Learning

Agriculture is considered as the backbone in India, where it supports about 18% of the total country GDP. Seed is the starting point of any kind of cultivation of plants, as they go through multiple stages to become as a plant. The seed are presented to the farmers through government and private agencies for different kinds of plants. There are various types of seeds such as organic, natural and hybrid, which brought by the farmers based on their requirement. Seeds are first tested and checked as a sample by the office of seed certification of the different states governments. The seeds are bought on lots or a certain required quantity by the farmers or public. The testing of the quality of the seeds are generally done through manual process, which is done as samples where there may be few wasted seeds in the packets or the lots purchased. This paper proposes a new quality checking process through machine vision system with deep learning, as the seeds are passed through the cameras and utilizing the image process technique with deep learning algorithms to match the quality which is trained to the system to identify and classify the seeds. Two different algorithms are proposed, U-Net and Fully Convolutional Regression Networks (FCRN) to classify and segregate the seeds. A mobile application has also been proposed and developed to experiment the process, as the application can be directly accessed by the public/farmers. As experimented with the algorithms, the proposed U-Net algorithm has shown better results on non-destructive detection with great potential for seed quality control and seed count.