Triple Sensors Integrated Deep Learning-Based Network for Detection of Defects in Medicines

The accurate medicine dispensing is extremely important in all hospitals, as medical errors can cause serious health issues, underscoring the need for improved protocols. Numerous research has been conducted on developing detection techniques for unpackaged pills or drugs with similar packages. However, the existing frameworks face issues, such as dispensing errors and accurate defect prediction. To overcome these challenges, a novel Tri-SID Medi technique has been proposed for detecting the defects in medicines and delivering defect-free medicines using tri sensors, such as ultrasonic sensor, weight sensor, and global positioning system (GPS). Initially, the collected drug images from the hospital/pharmacy are preprocessed using the adaptive unsharp mask-guided filter to enhance the image quality and reduce noise artifacts. Then, the preprocessed images are fed into APM-YOLO to detect the images as defect or nondefect. The nondefect medicines are given into the first similarity group (SG), in which the squeeze and excitation-based bidirectional long short-term memory (SE-BiLSTM) framework is trained for grouping the medicinal products. Finally, the defect-free group medicines are directly delivery to the customers in the delivery phase by using tri sensors and global system for mobile communications (GSMs). The proposed Tri-SID Medi method achieves 99.03% accuracy, which is 9.11%, 2.45%, and 5.07% better than the existing techniques, such as YOLO, deep convolutional neural network (DCNN), and bidirectional encoder representations from transformer (BERT). The experimental results of the proposed techniques highlight the potential for real-time drug defect detection and similar drug groupings with diverse packaging, which can reduce medical errors and are key to ensure patient safety.