In the field of fruit production, crucial decisions on crop management are influenced by bloom intensity, i.e., the number of flowers existing in an orchard. In spite of being significant, the estimation of bloom intensity is still carried out by humans’ visual inspecting it. The available residual Convolutional Neural Network (CNN) systems for flower identification can operate under particular conditions and offers a performance that does not reach the mark. In order to resolve this problem, CNN based fine-tuned via the new classifier model. A Weight Based Convolutional Neural Network (WECNN) is proposed that makes use of a ensemble method for improving the fruit flower detection model for CNN optimization. The enhanced model is selected on the basis of the fruit flower detection rate and the difference observed in the results of classification in comparison with classical CNN. For the improvement of the results, a confidence classification is constructed and a confidence function is designed on the basis of the regions to attain the CNN’s classification quality. WECNN needs to make coarse segmentations; a refinement technique is employed for differentiating between various flower instances in a better manner. In order to improve its sensitivity towards flowers, this network is fine-tuned employing a dataset of AppleA flower images. The results are then measured in terms of the metrics of precision, recall, f-measure and accuracy.
Volume 11 | 11-Special Issue