Deep Convolutional Neural Network for Multi-Class Flower Image Classification Using the Flowers Recognition Dataset

Abstract

Automatic flower species identification plays a crucial role in agriculture, horticulture, and biodiversity conservation, as manual identification is time-consuming and susceptible to errors caused by high intra-class variability and inter-class visual similarity. This study proposes a Convolutional Neural Network (CNN)-based approach for classifying three flower species—sunflower, daisy, and tulip—using the publicly available Flowers Recognition dataset from Kaggle, comprising 2,481 images. The dataset was divided into training (70%), validation (15%), and testing (15%) subsets. The preprocessing pipeline included image resizing to 224 × 224 pixels, pixel value normalization to the range of 0–1, contrast enhancement, and data augmentation through random rotation (±20°), horizontal flipping, zooming, and shearing to improve model generalization. The proposed CNN architecture consists of four convolutional blocks with 32–256 filters, followed by max-pooling, batch normalization, a dropout layer (0.5), fully connected layers, and a Softmax output layer. The model was optimized using the Adam optimizer with categorical cross-entropy loss and trained for up to 50 epochs using Early Stopping and ReduceLROnPlateau callbacks. Experimental results demonstrate that the proposed model achieved a test accuracy of 90%, with 90% precision, 90% recall, and a 90% F1-score. The sunflower class exhibited the highest recall (93%), while most misclassifications occurred between the daisy and tulip classes due to their similar visual characteristics. Furthermore, the regularization strategy effectively mitigated overfitting, as indicated by the relatively small gap between training and validation accuracy (97% and 90%, respectively). These findings demonstrate the effectiveness of CNNs for automated flower image classification and highlight their potential for supporting intelligent flower recognition systems. Future work will focus on expanding the number of flower categories and integrating transfer learning techniques to further improve classification performance.