Automatic classification of cowpea leaves using deep convolutional neural network

This paper uses a binary classification to discriminate cowpea leaves from other/weed leaves. Cowpea is one of the nutritional crops, but research on cowpea is negligible worldwide. Hence our experiments are purely based on cowpea leaves to increase the productivity of the high fiber consumed. An automatic model implementation has yet to be done on the cowpea leaves dataset. Therefore, our work is essential for researchers in this field. In such cases, various state-of-the-art algorithms are available, but our approach appears competitive against existing ones. The overall analysis of the complete results we have cited in the paper in table format. The images have been collected from the cowpea field of The Indian Council of Agricultural Research (ICAR), New Delhi and dataset constructed in the lab. The plants are grown under suitable environmental conditions, and images have been collected with a standard DSLR camera. In the proposed CNNs model, data (images) of cowpea leaves have been collected from a Research farm. For all collected data, we have made labeled data (Databank) to utilize cowpea leaves for further research, and then we applied three convolutional neural networks to classify cowpea leaves for smart agriculture. The experimental results show that the DenseNet121 realizes a detection performance with the highest accuracy on the CLDC. We have also used two more CNN architectures to identify cowpea leaves from the weed, and a comparative study has been made and explored in the paper. DenseNet121 methods give an accuracy of 86.12% (training dataset) and 88.89% (testing dataset), respectively.