Embedded Machine Learning for Mango Classification Using Image Processing and Support Vector Machine

Agriculture is an important economic aspect of many Asian countries, including Vietnam. One major obstacle, which prevents Vietnam from exporting some fruit species to western countries such as mangoes, is the irregular quality of the harvested fruits. Various research aimed to solve this problem by proposing different methods of estimating the weights of the fruits to determine anomalies. However, most methods are only tested for their maximum accuracy under favorable conditions with expensive equipments, and are therefore often not suitable for the poor farming regions of the developing countries. In this paper, with the main focus being mangoes, we would like to study a simpler approach using a combination of images processing techniques and support vector machine (SVM) for determining the approximate weight range of the fruits. The target device, on which the approach will be tested, is a Raspberry Pi 3B, which is fairly affordable even in developing countries. This will examine the feasibility of the weight classification for agriculture using only cheap hardware.