AgroAVNET for crops and weeds classification: A step forward in automatic farming

Convolutional Neural networks have endeavored to solve various problems in different fields such as industries, medication, automation, etc. Among these areas, automatic farming is one of the important application and crop management is its most crucial part. It is necessary to recognize weeds in an early growth stage so as to control their side effects on the growth of crops and increase the yield. This work is an attempt to classify weed and crop species by using convolutional neural networks. To achieve this, AgroAVNET which is a hybrid model of AlexNet and VGGNET is proposed. Its performance is compared with AlexNet, VGGNET and their variants and existing methods for crop-weed species classification. This work also deals with how an existing system can be used to learn new categories of weeds and crops. Plant seedlings dataset is used for evaluation of the proposed system. Average accuracy, precision, recall and F1-score are used as performance metrics. It is seen from experimental results that, AgroAVNET outperforms AlexNet and VGGNET. Also, it takes less training time to learn new species compared to scratch training.