Filters
Convolutional Neural Network Model for Variety Classification and Seed Quality Assessment of Winter Rapeseed
The main objective of this study is to develop an automatic classification model for winter rapeseed varieties, to assess seed maturity and damage based on seed colour using a convolutional neural network (CNN). A CNN with a fixed architecture was built, consisting of an alternating arrangement of five classes Conv2D, MaxPooling2D and Dropout, for which a computational algorithm was developed in the Python 3.9 programming language, creating six models depending on the type of input data. Seeds of three winter rapeseed varieties were used for the research. Each imaged sample was 20.000 g. For each variety, 125 weight groups of 20 samples were prepared, with the weight of damaged or immature seeds increasing by 0.161 g. Each of the 20 samples in each weight group was marked by a different seed distribution. The accuracy of the models’ validation ranged from 80.20 to 85.60%, with an average of 82.50%. Higher accuracy was obtained when classifying mature seed varieties (average of 84.24%) than when classifying the degree of maturity (average of 80.76%). It can be stated that classifying such fine seeds as rapeseed seeds is a complex process, creating major problems and constraints, as there is a distinct distribution of seeds belonging to the same weight groups, which causes the CNN model to treat them as different.
Convolutional neural network model for the qualitative evaluation of geometric shape of carrot root
The main objective of the study is the development of an automatic carrot root classification model, marked as CR-NET, with the use of a Convolutional Neural Network (CNN). CNN with a constant architecture was built, consistingof an alternating arrangement of five Conv2D, MaxPooling2D and Dropout classes, for which in the Python 3.9 programming language a calculation algorithm was developed. It was found that the classification process of the carrot root images was carried out with an accuracy of 89.06%, meaning that 50 images were misclassified. The highest number of 21 erroneously classified photographs were from the extra class, of which 15 to the first class, thus not resulting in significant loss. However, assuming the number of refuse as the classification basis, the model accuracy greatly increases to 98.69%, as only 6 photographs were erroneously assigned.