Inkjet-printed fully customizable and low-cost electrodes matrix for gesture recognition

Abstract

The use of surface electromyography (sEMG) is rapidly spreading, from robotic prostheses and muscle computer interfaces to rehabilitation devices controlled by residual muscular activities. In this context, sEMG-based gesture recognition plays an enabling role in controlling prosthetics and devices in real-life settings. Our work aimed at developing a low-cost, print-and-play platform to acquire and analyse sEMG signals that can be arranged in a fully customized way, depending on the application and the users' needs. We produced 8-channel sEMG matrices to measure the muscular activity of the forearm using innovative nanoparticle-based inks to print the sensors embedded into each matrix using a commercial inkjet printer. Then, we acquired the multi-channel sEMG data from 12 participants while repeatedly performing twelve standard finger movements (six extensions and six flexions). Our results showed that inkjet printing-based sEMG signals ensured significant similarity values across repetitions in every participant, a large enough difference between movements (dissimilarity index above 0.2), and an overall classification accuracy of 93-95% for flexion and extension, respectively.

Figure 1

Representative bubble plot showing the signal quality during the flexion movement of the middle finger in a single participant. Each bubble shows the average normalized RMS value across repetitions of the same movement (color code) and its standard deviation (diameter). The corresponding average sEMG signal across trials is shown for each unipolar channel. Figure created with MatLab 2020a and Microsoft PowerPoint 2016.

Figure 2

Intra-subject similarity analysis for the flexion (F) and extension (E) tasks. The similarity coefficient (averaged across repetitions) is shown for all the gestures (color code), and subjects, for the flexion (upper panel) and extension (lower panel) sessions, separately. Note that the y-axis ranges from 0.3 to 1. Figure created with Matlab 2020a.

Figure 3

Intra-subject variability analysis for the flexion (F) and extension (E) tasks. The dissimilarity coefficient between pairs of gestures is shown in every subject, for the flexion (upper panel) and the extension (lower panel) sessions, separately. Values below 0.2 are not represented. Figure created with MatLab2020a and Microsoft PowerPoint 2016.

Figure 4

Confusion matrices were obtained in the multi-class classification problem for the flexion (F) and extension (E) tasks. Flexion (upper panel) and extension (lower panel) gestures were analyzed separately. From left to right, DA, SVM, and kNN results are shown. The average accuracy value for each of them is as follows: Flexion: 93% for DA, 83% for SVM, 82% for kNN; Extension: 95% for DA, 90% for SVM, 72% for kNN. Figure created with MatLab2020a and Microsoft PowerPoint 2016.

Figure 5

Graphical representation of the fabrication process of the electrodes’ matrix and their positioning on the forearm. (a) The Epson Stylus 1500 W consumer printer used for the printed sensors, loaded with AgNP and standard black ink. (b) SEM image of the printed AgNPs. (c) CraftROBO C330 digital blade cutter for the patterning of the biadhesive sheet. (d) AutoCAD matrix design with cutting lines and mirrored text to identify the electrodes through the transparent PET film after its application on the skin. (e) Printed electrodes’ matrix with the biadhesive passivating layer sticked on top. (f) Matrix of electrodes gelled and ready to be positioned on the participant forearm. (g) Example of the matrix positioning for extension tasks and of the readout of the signals between the electrodes, forming the 8 channels (arm original drawings courtesy of Maria Catalina Li Puma). (h) Readout channels positions for extension (right) and flexion (left), (arm original drawings courtesy of Maria Catalina Li Puma). (i) Matrix positioned on the forearm of one of the subjects. Figure created with Microsoft PowerPoint 2016.

Figure 6

16 gestures performed by the 12 participants, randomization of their performance and structure of each acquisition of each gesture. The last two gestures in each row are the same and have been used as control movements. (a) Visual description of the gestures and of their execution order (depending on the subject group). To each gesture, it has been assigned a univocal code composed by a letter (F for flexion and E for Extension) and a progressive number. (b) Table identifying the muscle group and task execution order for each participant. (c) schematic of the acquisition performance for each gesture. Figure created with Microsoft PowerPoint 2016.