Enriching the Multi-Object Detection using Convolutional Neural Network in Macro-Image

An object recognition and localization is a primary issue that is harder than a classification of an image even with precise object location and their annotations available at the time of training. The feature is identified for localizing the objects and classification identifies the classes from recognized object regions. This research work is proposed the two approaches i. Support Vector Machine (SVM) is optimized using the Firefly Algorithm (FA), and Scale Invariant Feature Transform (SIFT) descriptors, ii. Convolutional Neural Network (CNN) with Adam (Adaptive Moment) optimizer. In first method, FA has been used in the searching of optimal parameters through the simulation of the social behavior of the fireflies using the bioluminescent i.e. emission of light intensity. This FA is trained the Lagrangian multiplier and smoothness parameters in the SVM continuously. The second research work Adam based CNN has recognized the objects in multi-object images. The hash directory is proposed to store the highest scored bounding boxes to speed up the process. The experiments have been evaluated with binary as well as multi-object images. The proposed model has been trained and validated using VOC2012 dataset. It consists the kind of general macro photography images for object identification. The Average Precision (AP) is computed, and comparison shows that CNN performs better than FA-SVM method.