Fractional deer hunting optimization algorithm enabled deep learning framework for sugarcane billet damage classification

As a significant raw material for the production of sugar, ethanol, and bagasse, sugarcane is a frequently used crop in agriculture. Machines are used almost everywhere to harvest the sugarcane crop. Today, billets are used to cultivate sugarcane. Thus, for efficient cultivation and maximum production, it is essential to have healthy billets. However, harvesting equipment might harm the billets and reduce their quality. Moreover, damage in the billets easily spreads the disease. Thus, there is a need for efficient techniques for sugarcane harvesters in order to detect the damage of sugarcane billet in a dynamic environmental condition. However, the conventional methods lack accuracy because of the complex background. Therefore, this research develops an efficacious model for sugarcane billet damage categorization by employing a fractional deer hunting optimization algorithm-based deep maxout network (FDHOA-based DMN). Initially, the median filter is employed to pre-process the acquired image and then the pyramid scene parsing network (PSP-Net) is utilized for the segmentation process. Moreover, the deep maxout network (DMN) is devised to classify the damaged sugarcane billet and the DMN is trained using the FDHOA. The reason for the selection of DMN for the classification process is that it has efficient learning capability and identifies the intrinsic features from the data. The fractional calculus and the deer hunting optimization algorithm are unified to obtain the devised FDHOA. In addition, the testing accuracy, sensitivity, specificity, and F1-score metrics are used to evlaute the efficiency of the devised model and obtained 0.940, 0.930, 0.959, and 0.935, respectively.