Multi-layered CNN Architecture for Assistance in Smart Farming

Extensive use of herbicides for weed management poses a major threat to the environment and food supply chain. The extent weeds take away the resources of the crop desired depends on the nature of the soil along with the type of the crop grown. Artificial intelligence enabled technologies are replacing human involvement in either removing the weed or targeted spray of the herbicides. However, the highest degree of similarity between weed and the desired crop as well as the existence of wide range of weeds affect the classification accuracy. Lack of ample labelled data is another problem there by affecting the overall problem of weed detection. These issues can be addressed by pooling up larger datasets under different conditions followed by making use of relevant advanced algorithms. Current work proposes a multi-layeredCNN model for identifying the weeds in Capsicum Annuum (Chilli) particularly in red sandy fields. The crop dataset was acquired under different growth stages of the crop and fed to the model along with different varieties of weed data set. Promising results were obtained with an accuracy of 99.92% within lesser number of epochs.