Block Level Optimization Based Mapping and Ensemble Classification Based Segmentation for Fruit Flower Detection

Abstract

In the field of agricultural, the crop estimation task entirely depends on the process of detecting and counting the number of fruits on trees. The crop estimation task is mainly used in the agricultural field. Currently, counting the fruits and vegetables manually is performed at several locations. Manual counting has several disadvantages since it consumes too much time and needs an excessive amount of labor. The agriculture productivity can be improved by incorporating the automatic fruit counting technique with a crop management system to gather the information to predict the yield. This information may be used to schedule the harvesting. This research approach creates maps about flowers at the block level. This research work employed the Modified Bat Optimization technique for the generation of a direct map from blocks with no computationally intensive refinement phase. Once the block-level mapping is done, then the ensemble classification is introduced for flower identification, which is reliable against uncertain environments and suitable for various flower species. The proposed technique depends on the Support Vector Machine (SVM) classifier and an Enhanced Convolutional Neural Network (ECNN) for carrying out semantic segmentation. To improve its sensibility towards flowers, this Ensemble Classification Framework (ECF) is fine-tuned with the help of a weighted majority function of apple flower images. Also, a refinement technique is used for all the classifiers to better differentiate between different flower instances. The dataset with pixel-accurate labeling is used for the implementation of the proposed technique. The dataset has images with high resolution.