Corn Seed Classification Using Deep Learning as an Effort to Increase Corn Productivity

Corn has an important meaning in the development of the agricultural industry because it is one of the raw materials for the food industry and animal feed industry. With the development of the food processing industry, the demand for corn has increased. There is one important obstacle, namely the selection of the quality of corn seeds so that the quality of corn produced is maximized. To overcome these obstacles, one of them is machine-based learning. The aim is to perform automatic classification of the seed quality of corn seeds. One method that can be used is the convolutional neural network because this algorithm has very good classification accuracy results, but the weakness is that the training time is quite high. The architecture chosen is the residual network and custom layer. Improvements made to the method are the use of a custom layer to reduce training time, the addition of a proposal region to focus more on corn seeds, and augmentation to increase the variation in the amount of data. The results showed that the results of the accuracy of the pattern recognition of corn seeds produced on average were 99,3% with the computation time of the training process with a custom layer of 2 minutes 32 seconds. The Results of average classification error for both architectures are Mean Square Error 0,0080, Root Mean Square Error 0,0935, and Mean Absolute Error 0,0063.