Broccoli Classification: A Fusion of CNN and AdaBoost
Arshleen Kaur, Rishabh Sharma, Nitin Thapliyal, Manika Manwal · 2024
Although precision farming has been introduced, differentiation of crop species still calls for the development of quick and reliable classification tools used in the process of optimizing cultivation practices, which ultimately boost crop yields. Here we propose a new classification method to identify broccoli varieties particularly of ‘Pusa KTS-1’, ‘Green Magic’, and ‘Calabrese’ with the use of an integrated approach that fuses the strength of Convolutional Neural Networks (CNN) and AdaBoost. This research makes a striking improvement in the accuracy of crop variety classification using the fusion of CNN for powerful feature extraction and AdaBoost which leverages the hard instances to increase the accuracy of the classifier. A database of broccoli images was prepared, auxiliary operations were performed and its diversity was improved to make the model not weak and generalized enough. On this dataset, the CNN-AdaBoost ensemble was trained, validated, and tested and the model achieved an overall accuracy of 93.7% which outperformed standalone models. The model's performance at boundaries validates its potential for automatic classification, which is important for precision agriculture's targets of higher productivity and efficiency. The results of this research consequently transcend beyond the scope of broccoli variety classification, lending ways for improvements in agricultural technologies through the adoption of machine learning. In the future, the model will be applied to other crops and will be used to solve many agricultural problems. Its parameters will be improved to make it more convenient for daily use.