In the Indian-origin Kashmir valley, apple trees are among the most popular plants. Thousands of tons of apples are exported from Kashmir every year, generating substantial revenue. Many diseases affect apple trees, which results in devastated apple yields and major losses for apple growers. Apple plants are mostly infected with diseases that originate in the leaves. In a country like India, where half of the population works in agriculture, prompt detection and prediction of such diseases is critical. Traditionally, apple plant diseases were diagnosed using laboratory assistance, which is time-consuming and labor-intensive. This study began with the construction of an expert-annotated dataset of apple diseases of suitable size, containing around 10000 high quality RGB images covering all of the main foliar diseases as well as their symptoms and signs. As a next step, we propose the development of a deep learning-based apple disease detection system that can be used to identify symptoms efficiently and accurately. Today, machine learning and deep learning allow us to reliably detect whether a plant is infected or not. This article introduces a framework for predicting Kashmiri apple plant diseases based on deep learning. Our model relies on Convolutional Neural Networks (CNNs) to extract and predict features. Testing samples through our framework yields state-of-the-art results in identifying apple plant diseases with 92 percent accuracy. Further, we present a novel dataset containing samples of Kashmiri apple plant leaves with three distinct diseases. A comparison was made between five deep learning algorithms: Faster R-CNN, Tiny-YOLO v2, YOLO v3, SSD 300, and EfficientDet-D0. The purpose of this was to demonstrate the efficacy of the method. As a result of the comparison, the recommended method outperformed the other five algorithms in terms of mAP (98.21%), 75.34 f/s detection speed, 14.48 MB model size, 13.27%, 18.53%, 16.19%, 5.17%, and 7.89%. Additionally, it meets real-time requirements. Furthermore, various lighting conditions and apple tree varieties were studied in order to identify apple blooms. Based on the information, it was demonstrated how robust the approach was against different varieties of apple trees and lighting patterns. This method demonstrated the efficacy of real-time apple bloom detection. Orchard yield estimates and apple blossom thinning robots may benefit from these findings.